StereoCloud3D User Guide

StereoCloud3D is a Windows desktop application for stereoscopic restitution: it lets you view massive LiDAR / photogrammetric point clouds and digitize accurate 2D/3D CAD vector features directly on top of them, in true 3D stereo when you have the hardware for it.

Who this guide is for

This manual serves two audiences. Operators will find step-by-step workflows, keyboard shortcuts and reference tables. Decision makers / prospective customers can read the Overview and the feature chapters to understand what the product does and how it fits a survey, mapping or AEC pipeline.

2. Key concepts

3. System requirements

• Operating system: Windows 10 or Windows 11 (64-bit).
• GPU: an OpenGL-capable graphics card. A discrete GPU is strongly recommended for large clouds and stereo.
• Memory: 16 GB minimum; 32 GB or more for very large datasets.
• Disk: fast SSD; the converted tile index is stored alongside your project.
• Antivirus / Windows Defender: real-time scanning of thousands of tile files during cloud import can slow conversion dramatically. The installer offers to add recommended Defender exclusions (app folder + converter processes). For projects stored outside the install folder, also exclude your project root(s) in Windows Security → Virus & threat protection → Manage settings → Exclusions, or run configure-defender-exclusions.ps1 from the install directory with -ProjectRoots. Third-party antivirus products need equivalent exclusions manually.
• Stereo hardware (optional): a 3D-capable display or projector for side-by-side output (e.g. PluraView / VoxelSpace / VIPTA StereoVisor side-by-side, or PluraView triple with one eye per panel).
• 3D navigation (strongly recommended): a trackball used together with a normal mouse. Primary (best fit): Kensington Expert Mouse or Expert Trackball (USB 047D) — large ball, scroll ring, four buttons. Also supported: Logitech ERGO M575 (current thumb trackball), and legacy Wireless Trackball M570 / Trackman Marble (USB 046D, often refurbished/used). Ball = digitizing Z quota; ring/wheel = Z speed; mouse MMB + ball = cloud rotation; MMB drag = pan. See Trackball navigation.
• Internet: required at activation and periodically for license validation.

4. Activation & licensing

At every launch StereoCloud3D shows the Activation window (StereoCloud3D — Activation). Enter the license key supplied with your subscription and click Activate.

• Activate — validate the key online and bind this computer to the license.
• Quit — close the application without activating.

You only type the key once. After the server accepts a key, the next launch opens with that key already in the field and the cursor in it: press Enter (or click Activate) and you are in. To use a different license, select the text and type over it.

While running, the app re-checks the license periodically. If the license expires or is revoked (beyond any grace period) the application will notify you and close. See Managing your license for the status dialog and how to move a license to another computer.

Update notification. After activation the app checks (in the background) whether a newer version has been published. When one is available, an Update available panel shows the new version number together with a concise list of the additions and changes, and a Download… button that opens the download page. Choose Later to dismiss it — the notice reappears at the next launch until you update.

5. Welcome screen

When no project is open you see the Welcome screen. From here you can:

• Open Project… — pick an existing project.
• New Project… — create a fresh project (see below).
• Exit — close the program, the same command as File → Exit. It sits beside the two buttons above so that changing your mind on this page needs no menu.
• Recent projects — a wider list with name, path, short summary and a thumbnail of the project. Missing projects are shown in orange; entries without a preview show No preview.

The thumbnail. It is taken by the program the first time you open the project, and only if the project has none yet: opening it again does not retake it. The program waits until the cloud of the view is completely loaded (at most 20 seconds), then keeps the part of the picture that actually contains data — with a margin, in the proportions of the card — so that a narrow strip of survey is not reduced to a thin dark line on a black card. When the view is not looking from above, the picture is taken after about 3 seconds instead of waiting for a plan view. A view that comes out empty is retried a few times and then kept as it is.

Click the thumbnail to open the project — it does the same as the Open button of that entry (the pointer becomes a hand and the border lights up when you are over it).

Each recent entry offers Open, Protect… / Password… (manage a project password) and Delete… / Remove from list. Password-protected projects carry a locked badge. Deleting offers the choice to permanently remove the project from disk or merely drop it from the recent list.

You can return to this screen at any time with File → Home / Welcome (you will be prompted to save unsaved changes first).

6. Projects

6.1 Creating a project

A new project starts from the survey data you were given. Choose File → New Project… (Ctrl+N). The dialog has two screens: Where is the survey data? and Here is what I found. Nothing is copied or converted until you press Create project.

A new project starts in stereo when the last project you had open in this session was in stereo, and in mono otherwise; an existing project always keeps its own setting.

Screen 1 — Where is the survey data?

There is one place to give the data, whatever it is:

• Choose the survey folder… — the folder you were given. The program looks inside it (two levels down, up to 5000 entries) and finds what it holds by itself. A folder that is itself a StereoCloud3D project or an octree is not entered.
• or choose single files — point clouds, orthophotos and a drawing, selected together.
• Drag and drop — drop a folder or files on the program window while the dialog is open. It is the same as choosing them.
• Create an empty project — a project with no data; everything can be added later.

If the folder holds no survey data the dialog says so (I found no survey data in this folder, with the number of files it checked). If it holds more than 5000 entries it says it looked only at the first ones and asks for a more specific folder. If it finds several mobile runs it asks you to choose the folder of a single run (Choose the run…): a project takes one run.

How the program sorts the files

Every file ends in exactly one place: used, or listed as not used with the reason. The program sorts by type:

DWG files. The program imports DXF; a DWG you add is not converted. Export the drawing as DXF from your CAD program.

Screen 2 — Here is what I found

The second screen is a summary, with the name and the folder already filled in. From top to bottom:

• The verdict. One line, Acquisition: Aerial / Terrestrial / Mobile, with the reason (for example trajectory of 612 stations with photos). A guess says (probably). See 8.7.1 for how it is decided. For photogrammetry the line adds a note that only a surface is probably in the cloud, so wires and the ground under trees may be missing. It is information; no tool is switched off.
• What was found, one row per kind: Mobile run (name, stations, clouds), Point clouds (files and total points when every header could be read), Orthophotos, Drawing, and Not used (hover to see which files and why). A small cross on a row removes that group from the list. Nothing on disk is touched.
• Several cloud files. You choose once: one layer (recommended) or one layer per file. One layer thins the overlap between pieces of the same survey and navigates faster; one layer per file keeps each file separate, to be switched on and off by itself. The choice is remembered.
• What do you want to produce? The job types, below.
• Name and It will be saved in (Change…). The name is proposed from the data. If a project of that name already exists in the folder, the program proposes the next free name and says so; a name you typed yourself is never changed. The last folder used is remembered; the first time it is Documents\StereoCloud3D.
• Options (closed unless you opened it; it stays as you left it), then Back and Create project.

What do you want to produce?

You choose what you want to produce, not how the survey was made. The list is the job types shipped with the program, as check boxes: tick one or more. A job type that these data cannot support is greyed, and its tooltip says why (Not with these data. Survey it is made for: …). The first one you tick is where the project starts, and is marked (start) once you tick a second one. The others are added to the project and you can switch to them later (8.7). Create project stays off until at least one is ticked, when there are data and catalogs to choose from. If no job catalog is found next to the program, the dialog says so and the project is created without one.

Options

• Workspace. The kind of survey is read from the data, not chosen. If the program got it wrong, correct it here with Change (Aerial, Terrestrial or Mobile). Without a recognised survey type, the workspace follows the first job type you ticked.
• The files, as lists. Point clouds and orthophotos, each with a cross to take one out, and Add more files… / Add another folder…. A cloud that is both listed on its own and inside the mobile run is imported once, and the dialog says so.
• The run's detail for a mobile run: its tracks, the photograph coverage, the clouds against the stations. It opens by itself once when the run needs a look.

Checks before Create

The reason a button is off is written above it. The dialog checks, and says in words:

• Conflicting coordinate systems. If two clouds are in different systems the program does not convert: export them again in the same system, or create two projects. The same applies in Add data to a cloud that differs from the project's system.
• A name reserved by Windows (CON, NUL, COM1…), a name with characters that cannot be used (\ / : * ? " < > |), or ending in a dot or a space.
• A destination you cannot write to: Choose a folder of yours, for example in Documents.
• A destination that already holds a non-empty folder of that name.
• A drawing without a cloud. The drawing sits on the cloud: with no cloud in the project it is not imported. The dialog says so before Create, and again at the end if it was skipped.
• More than one mobile run, and a run folder that cannot be read.
• An import still running from a previous request.

Create project

Create project makes the project folder, then imports the data one after the other — the run, the clouds, the images, the drawing last — each with its own progress. At the end the bottom panel opens (see 8.7.7). A project with the drawing but no cloud still works; the drawing is skipped and the status line tells you to import it again once the cloud is loaded.

• Cancelling a cloud import. The import window has Cancel import. After you confirm, the converter is stopped and the partial files are removed. The project stays, empty of that cloud, and nothing else of the plan runs. Clouds already in the project are not touched. Add the clouds again with File → Add data….
• When an import is refused or fails the dialog says why and the project says that nothing was added: no active project, nothing to import, another operation on the data still running, the disk not reachable, the cloud converter missing next to the program (reinstall), clouds in different coordinate systems, or a licence that does not allow importing clouds. A failure shows the converter's own message. There is no silent “done”.
• If the project folder could not be made or written (path not valid, folder not created, project file not written, folder incomplete after creation), the project is not created and the dialog names the folder and the likely cause.

Add data to an open project

File → Add data… (with a project open) opens the same two screens without the name, folder and job-type parts: Add survey data — where is the data, then what the program found, with the same sorting, options and checks. The button is Add and import; it imports the list into this project, one item after the other, each with its progress, and opens the bottom panel at the end. It stays off with Add at least one file or a run folder first until there is something in the list.

6.2 Password protection

A project password is a soft access gate: when enabled, opening the project requires the password and an operator name (recorded in the job report). You can set, change or remove a password from three places:

• File → Project Security… — Enable / Update / Remove (removing requires the current password).
• Welcome → Protect… on a recent project.
• File → Job Report… → Set password… (Security: protected… once set).

The New Project dialog no longer asks for a password (or a drawing template): a project is created unprotected, and you protect it afterwards from one of the places above. Opening a project that already has a password works as before: the program asks for your name and the password.

6.3 Saving

• Save (Ctrl+S) writes the project manifest and drawing.
• Save As… (Ctrl+Shift+S) writes a copy.
• The title bar shows the project name with a * when there are unsaved changes.
• Each explicit Save or Save As also creates a versioned CAD backup of the current master drawing before it is overwritten (see 6.5).

6.4 Autosave & crash recovery

Autosave is on by default at a 5-minute interval (configurable 1–240 minutes in the Inspector panel). Autosave writes to a sidecar file (*.autosave.vec) — it does not overwrite your master drawing until you explicitly save. If the app finds a newer sidecar when opening a project it offers to Recover or Discard the unsaved changes. When closing with unsaved edits you are asked to Save, Don't save or Cancel.

Two recovery levels. The autosave sidecar protects against crashes and unexpected shutdowns. The rotating backups created on Save protect against mistakes after you have already committed a good drawing to disk (see below).

6.5 CAD drawing backups & restore

Release 3.20 adds a second backup level for the native CAD drawing:

• Level 1 — Autosave sidecar (cad/drawing.autosave.vec): written periodically while you work; offered automatically on reopen after a crash.
• Level 2 — Versioned backups on save: every time you choose Save or Save As, StereoCloud3D copies the current master cad/drawing.vec into cad/backups/ as drawing_YYYYMMDD-HHMMSS.vec before overwriting the master. The ten most recent snapshots are kept; older ones are pruned automatically.

To reload an older drawing without hunting through folders, use File → Restore from backup… (Ctrl+Alt+R). The picker lists both autosave sidecars and timestamped backup versions (newest first), with file size and date. Choosing a backup reloads that .vec into the session and marks the project dirty until you save again — your current master file on disk is not touched until you explicitly save over it.

7. Importing data

7.1 Importing point clouds (LAS / LAZ / E57)

Open the Project Manager (Ctrl+P) and choose Import cloud…. Select one or more .las / .laz / .e57 files (multi-select is supported).

E57 (terrestrial scanners). .e57 is the format Leica, FARO, NavVis and Trimble export. One file can hold several scan stations, and each station stores its points in the scanner’s own frame together with the pose that places it in the survey: StereoCloud3D applies that pose, so stations arrive where the registration put them instead of piled on one another. Colour and intensity are read when the export carries them, invalid points are discarded, and the coordinate system is taken from the file’s coordinateMetadata when it declares one — many exports leave it empty, and the import then reports “no coordinate system declared” rather than guessing.

New imports are converted with the bundled octree_builder into a streaming octree under cloud/<layer>/octree/ (metadata + hierarchy + blobs). When you select several tiles of the same flight/block, StereoCloud3D asks whether to merge them into one cloud (one octree; overlapping tiles are decimated together) or keep one layer per file.

While conversion runs, a progress dialog (Importing clouds) reports:

• Dataset — number of files and total size on disk (MB).
• Overall progress across all files (i / N files).
• Current file name, its point count (or “reading header…”), and the current phase.
• A per-file progress bar.

Minimum point spacing

Two points closer together than the minimum point spacing are surplus from overlapping passes: the octree stops subdividing at that cell size and keeps one point per cell. The control sits in the Project Manager, above Import cloud…, and it is set before the import rather than discovered afterwards in the report.

• Measured on the data (recommended) — the converter measures the delivery's own native spacing (the median distance from a point to its nearest neighbour, in 3D) and uses it, never going above the ceiling of the current workspace: 2 cm for Mobile Mapping and Terrestrial Mapping, 5 cm for Aerial Mapping, 5 cm for a project with no workspace. The measured value is written into the import log.
• Set by me — the value you type is used exactly as given, with no ceiling.
• None — every point is kept.

A stray point far from the cloud (5.4)

One wrong return far from the cloud (a reflection, a bird, a corrupted record) used to stretch the whole octree: the origin and the root cube were chosen to contain it, so the levels were coarse for the real cloud. A cloud of 0.8 km came out with a cube of 292 km. Now the converter keeps the 4096 extreme points per axis and drops a stray when its gap from the rest is larger than the thickness of the rest — there is no fixed distance. The same cloud now gets a cube of 787 m. A notice names the axis, and the number of points dropped is written in the octree metadata (outlierPointCount). The points are not deleted from your data: they are only left out of the octree. To switch the filter off, run octree_builder with --no-outlier-filter.

Many clouds in one project (5.6)

A job made of many clouds stays in one project. Only the clouds that are on are loaded and drawn, together; a cloud that is off is unloaded (it uses no memory) and stays on screen as its outline, with a small Turn on button in the middle. Hover the outline to read the cloud’s name. Switching clouds on or off never moves the view.

• Project clouds bar (Cloud menu, also in Keyboard Shortcuts) — one box per cloud to switch it on and off, and Only to keep just that one. It opens by itself when the project has two clouds or more, and is remembered.
• Scene Tree, right-click on a cloud — Show only this cloud turns it on and the others off; the tick box switches the single cloud.
• A cloud imported later is added to the others and is drawn with them.
• The memory for decoded points is shared among the clouds that are on, so many clouds no longer add up to more memory than one.
• An outline drawn dashed is approximate and larger than the cloud: the import did not record the extent of the source files.

Add files to this cloud… (right-click on the cloud in the Scene Tree) rebuilds the cloud together with the files you choose, as one cloud, in the same import with progress and Cancel. The original files of the cloud are not needed and its classes are kept. Saved classification versions no longer apply to the new cloud, so none stays active. The previous cloud stays in the project folder as a copy (cloud/<name>__prima_<date>); if the rebuild fails, the project keeps the cloud it had.

Import report

When conversion succeeds an Import report summarizes the result:

• Layers imported, Total points, Total source size (MB).
• Extent (local) — width × depth and height of the combined bounds.
• Attributes present — XYZ always, plus RGB and/or Intensity when available.
• A per-layer table: layer name, type (LAS/LAZ), points and size.

If conversion fails, a Cloud import failed dialog shows the error message together with the tail of the converter's output to help diagnose the problem.

Project Manager — Properties tab

Release 3.20 adds a Properties tab to the Project Manager (Ctrl+P). It summarizes the open project without opening folders in Explorer:

• Storage — project path, total size on disk, created/modified timestamps from the manifest.
• Point clouds — layer count (visible/total), indexed point count, tile count, LAS source count and points, cloud data size on disk.
• Imagery — layer count (visible/total) and imagery folder size.
• CAD — drawing file name, file size, entity count (when CAD is loaded), CAD folder size, number of versioned backups and autosave sidecar presence.

Statistics are cached for a few seconds so the panel stays responsive on large projects.

7.2 Adding an existing tile index

If you already have a converted tile_index.json, use Project → Add cloud tile index… to attach it without re-converting.

7.3 Importing CAD (DXF / VEC)

Use Project → Import DXF / VEC… or the equivalent entry of the CAD menu. Imported geometry is converted into the project's native drawing. To send work back out, use CAD → Export drawing (DXF)….

• DXF — any version. The program reads DXF with its own reader and uses no external converter.
• VEC — native drawing format, copied in directly.
• DWG is not imported. A .dwg file is refused before any conversion, with the message “DWG is not imported: save the drawing as DXF (any version) and import the DXF.” Open the drawing in your CAD program, save it as DXF (any version) and import that. The file dialogs of CAD import, external references and new-project data offer DXF and VEC only.

DWG export (5.6): the drawing is written as AutoCAD 2004 DWG by the bundled sc3d_dwg.exe, from the same DXF the export writes. Before the file takes its name the tool reads the DWG back and compares the entities type by type; any error is reported with its reason. DWG is offered in the export dialogs when sc3d_dwg.exe is installed next to the program.

DXF as the drawing shows it (5.4)

The DXF reader reads what the drawing shows, as a reference reader does, not only the geometry: line types (dashed and the others, with their pattern), layers that are off (they stay off), aligned texts, hatch patterns, and arcs (their direction and angles now come out right, also inside polylines). Export keeps blocks, ellipses, arcs inside polylines and line types. What cannot be written as such is listed in the log, entity by entity type, so nothing is lost silently. The progress bar moves smoothly during both.

When the project has a cloud, the first view frames the cloud, not the extent of the drawing. A DXF whose content lies far from the cloud (for instance tens of thousands of attribute blocks over tens of kilometres around a cloud of a few hundred metres) no longer leaves the cloud a few pixels wide in the middle of the screen. Without a cloud the first view frames the drawing, as before.

When an import, export or xref is refused

A drawing that cannot be read used to close the progress overlay exactly as a successful one did, leaving you to work out from the empty drawing that something had gone wrong. Now a dialog says what happened, and it distinguishes three cases:

• Refused — the file could not be imported. The dialog names the file and the reason. The most common one is a DWG file: DWG is not imported, so save the drawing as DXF (any version) and import the DXF.
• Imported, but empty — the conversion worked and produced no geometry. Usually a drawing whose content sits in an unsupported entity type, or in external references that were not supplied.
• CAD imported, with exclusions — the geometry is there, but some entity types were dropped. The dialog lists which, so you know what to look for before you start drawing on top of something that is missing.

7.4 Importing imagery / orthophotos

Release 3.18 adds a georeferenced imagery workflow; 3.71 adds ECW import (via a one-time QGIS LTR install), lists every layer in the Scene Tree, and shows visible imagery in the Zenith Overview as soon as tiles are ready.

Ways to add an image:

• File → Import imagery… (Ctrl+Shift+Alt+U), or type IMAGERY / ECW on the command line.
• Project Manager (Ctrl+P) → Imagery tab → Import imagery….
• Scene Tree — right-click Imagery (or a layer) → Add imagery… (opens Project Manager on the Imagery tab).

The importer converts the source to tiled GeoTIFF assets stored inside the project under imagery/layer_N/, plus an imagery_index.json for streaming.

• Georeferenced placement — imagery is read from its world coordinates and presented in the WCS / Zenith Overview below CAD vectors, so orthophotos and imported drawings can be checked together.
• Project-managed layers — each imagery import creates a project layer with a matching CAD layer entry. Showing/hiding the imagery layer (Scene Tree tick, Imagery toolbar, or Ctrl+Shift+I) keeps the CAD layer visibility synchronized.
• Runtime packaging — the installer includes the GDAL executables, runtime DLLs, GDAL data and PROJ data required by the imagery pipeline. ECW additionally needs QGIS LTR on the PC (see below).
• Diagnostics — if conversion fails, StereoCloud3D shows a detailed failure report with the converter output tail, so missing georeference or unsupported raster formats can be identified quickly.

Supported source formats

GeoTIFF/TIFF, ECW, JPEG, PNG and BMP. GeoTIFF and ECW carry their own georeferencing; JPEG, PNG and BMP need a world sidecar (.jgw, .pgw, .bpw, .wld) or an .aux.xml in the same folder. Whatever the source, the import stores a Cloud Optimized GeoTIFF inside the project, so the original format matters only at import time.

Coordinate system of the imagery

Imagery is never reprojected. The raster is drawn where its own coordinates say, so it lines up with the cloud only when both are in the same coordinate system. A raster in a different system still imports and still looks correct on its own — it simply lands beside the cloud, by the distance between the two systems.

StereoCloud3D compares the coordinate system declared by the raster with the one recorded for the project, and writes both into the log together with size, geotransform and extent. This runs twice: at the end of an import, tagged [imagery-import] and covering only the layers that import just added, and at every project open, tagged [imagery-open] and covering every imagery layer the project has — so a project imported before this check existed still puts its coordinate systems in a support log.

At import only, two outcomes also raise a dialog, because in both the photo may really be somewhere else:

• Different systems — the raster declares a coordinate system that is not the project's.
• No system at all — the raster carries only a geotransform (typically a world sidecar with no .prj), so its coordinates are taken as already being the project's.

The dialog names both systems and does not block anything: the import has finished and the layer is usable. Check the overlay against the cloud; if it is offset, re-export the raster in the project's coordinate system and import it again. When the two systems match, or when neither side declares enough to compare, nothing is shown and the verdict stays in the log.

Opening a project never raises the dialog, only the log lines. The question belongs to the moment the file is chosen and can be re-exported; asked again at every open it would become a warning the operator learns to close without reading.

7.5 Importing ground photographs

Project → Import photographs… copies photographs of the structure — the ones taken by hand from the ground, one per face — into the project and files them under Photos in the Scene Tree.

These are not imagery. An imagery layer is a georeferenced orthophoto draped on the world XY plane (§7.4); a ground photograph is a perspective picture of an elevation, taken from an unknown position. The two go through different paths on purpose.

• Where the files go — copied into <project>/photos, keeping your own file name, so the Scene Tree shows what the project actually contains and Delete can remove both the record and the file. The source folder is left untouched.
• Same name, different picture — a second photograph that happens to share a name is imported as name (2).jpg rather than overwriting the first. Importing the same file again lands on the same node instead of adding a second one.
• What is read — camera make and model, focal length, focal length as a 35 mm equivalent, date taken, picture size and orientation, exactly as the camera wrote them. Right-click the photograph → Details → Camera to see them. A picture that carries no EXIF — one that came through a chat application, or was re-saved by an editor — imports normally with those fields empty.
• Look at one — right-click → View image… opens it in the image viewer.

7.6 Orienting a photograph against a drawing

An imported photograph is stored and described, not positioned. Orienting it works out where the camera stood, by matching a handful of features between the picture and a drawing of the same face. Right-click the photograph in the Scene Tree → Orient against a drawing….

Two pictures appear side by side. Every pixel of the drawing already knows its position in space, so a click there is a 3D point — nothing has to be found in the 3D viewport.

• Choose the drawing at the top. Only drawings that have been generated can be paired: pairing needs their depth, and depth is produced by generating, not stored in the delivered image. Ungenerated drawings are listed but greyed out. Choosing one prepares it, which takes a few seconds.
• Click in pairs — the same detail on the photograph and on the drawing, in either order. The panel says which half it is waiting for. A pair completes on its own second click and the next one starts immediately; there is no button between them.
• Wheel zooms towards the cursor, middle-drag pans, double middle-click brings the whole picture back into view.
• Remove a point with the x on its row, or select the row and press Delete.
• Drawing brightness lifts a dark elevation so its features can be seen well enough to click. It affects the display only and never the geometry.

From the fourth point on, the position is worked out again after every change and the result appears above the list: the typical error and the worst point, each in pixels and in centimetres on the structure, plus how far away the camera was. Every matched point is numbered on both pictures and coloured by its own error — green under 2 px, amber to 6, red beyond — so a wrong pair can be seen rather than merely suspected.

The matched points and the resulting position are saved as soon as they change, in <project>/photos, and come back when the photograph is opened again. An oriented photograph also appears under Products → Oriented photographs with its drawing, its error in centimetres and its point count, because that error is what decides whether the position can be used.

8. The workspace

The main window is organised around a central 3D viewport with a menu bar on top, optional toolbars, and movable floating panels.

8.0 Desktop chrome (3.25)

With a project open, the mono content band uses a fixed left-to-right chrome layout:

1. CAD Tools — docked Draw / Modify strip (left).
2. Scene Tree — project hierarchy (clouds, AOIs, imagery, DTM).
3. Zenith Overview — optional plan map (when enabled).
4. Mono viewport — cloud + CAD (centre).
5. Object Snap — docked snap toggles (right).

Drag the vertical splitters between docks to resize; widths are remembered separately for each workspace (see 8.6, Your arrangement is remembered per workspace) once you have dragged one there, and fall back to your last width elsewhere until you do. Busy overlays dim only the mono viewport so Tools / Scene Tree / Overview / OSnap stay usable.

8.0.1 Where the job stands

At the head of the Scene dock, above the Project and Graph tabs, a strip names the workspace in its colour and lists the steps that kind of job goes through — for an aerial project: Data, Classification, Extraction, Drawing, Delivery.

Each step carries a mark read from the project itself:

• [x] — the project already contains what that step produces.
• [ ] — nothing that step produces is in the project yet.
• [?] — the project records nothing about that step, so its state cannot be read. Reviewing a station run and dimensioning a drawing are not written into the project, so those two steps always read this way. A mark that guessed would be worse than one that says so.

Nothing is stored to keep these marks: they are derived from the project every time it is drawn, so they cannot drift away from it, and a project made before this existed reads correctly without conversion. The steps orient the work — they never block it, and any of them can be done in any order.

Clicking a step selects, in the tree below, what that step produced. It moves the selection and nothing else: no panel opens, no view changes, no command starts.

8.0.2 Scene Tree

View → Scene Tree (shortcut in Keyboard Shortcuts) docks the scene graph beside the Tools strip. Nodes cover Project, CAD, Clouds, AOIs, Imagery, Photos and everything the job produced. Use the visibility checkbox to show/hide that segment. Only rows whose tick box actually switches something have one — clouds, imagery, ortho sheets, the surface models (DTM, DSM, nDSM) and any node that owns CAD entities; product family folders carry no tick box, because hiding a folder does not hide what is under it. Three rows are master switches for a whole kind of content rather than for a subtree: Clouds, CAD and Imagery. Switching Clouds off hides the point cloud and nothing else — a drawing filed under a cloud's processings stays on screen.

A product's name has to identify it, not classify it: surface products carry their mean elevation, and buildings carry their number, because six pitches of the same kind used to give six identical rows that said what they were and never which.

A folder opens on at most 50 rows. Past that it stops and offers Show the other N, which reveals the rest of that folder for the session. A job that numbers four hundred buildings would otherwise put four hundred leaves between you and the next branch. The choice is a way of looking, so it is not saved with the project.

The CAD row opens on the drawing's layers — those that hold at least one entity, with the count beside each name. Each tick box switches that layer on or off through the same native call the Layer Manager makes, so the two always agree. This is the axis that crosses all the buildings at once: switching FALDA off hides every roof pitch in the drawing, however and whenever it was drawn. The tick boxes on the Buildings products do something narrower — each one hides the entities that one reconstruction run created, and nothing drawn by hand belongs to any product at all. The surface models are an exclusive group: there is a single raster under the Zenith Overview, so ticking one unticks the others, the folder row says one at a time, and hovering the tick box says the same. The identical switch is in Display → Overview as radio buttons; both go through the same service, so the two can never disagree. When you open a project, the tree expands fully so every folder is visible.

Context menus (right-click):

• Cloud — color mode and cloud-related actions.
• Imagery (folder or layer) — Add imagery… opens Project Manager on the Imagery tab (GeoTIFF, ECW, JPEG, PNG). On a layer, View image… opens the GeoTIFF in the image viewer.
• AOI — define/redraw the rectangle that bounds where the algorithms compute, and Algorithms (Cleanup, Voxel, SOR) enabled according to the estimated point count (with downsample when over the process limit). The AOI does not limit what is displayed: the cloud is always drawn whole.

Selecting a defined AOI shows a short capability summary (estimated points, and which algorithms can run on them) at the bottom of the panel.

Products (elaborati)

Everything a job produces is filed under Products, grouped by family: surface models, buildings, spot heights, ortho sheets and their plans, elevations and sections, CAD sheets, exports and reports. A product carries the run that made it, its parameters, its status and the files it wrote, and its right-click menu offers Show in the Zenith Overview, Open in image viewer, Show files in Explorer and Remove from tree. A sheet that has already been exported or placed on a CAD sheet is marked delivered / on a sheet with the date.

8.0.3 Project Graph

Project Graph — read-only view of project data, processing and product dependencies. The Scene dock carries two tabs, Project and Graph: the first is the tree described above, the second draws the same model as a diagram. It is a view, not an editor. Nothing is created, renamed, deleted or connected from it, and no layout is stored: the picture is rebuilt from the project every time it is drawn, so it cannot disagree with the data it describes.

It describes the project and does not command it: there is no Show or Hide here. Selecting a node is not the same act as putting it on screen, and what goes on screen is chosen in the Display panel (8.0.4), which is the one place that governs representation.

Nodes are laid out in three fixed columns, left to right, so a dependency reads in the direction it flows:

• Data — what was loaded or imported: point clouds, orthophotos, the project CAD drawing.
• Process — a run that was executed, such as a classification.
• Product — what a run produced: surface models, buildings, spot heights, exports and the rest.

Each card carries two lines, with a coloured edge marking the category. The first names the node — for a run, the operation it performed. The second names the run it belongs to, as a compact local date and time: two executions of the same operation are otherwise indistinguishable, and a product shows the run that last wrote it rather than the day the node was created. A node that belongs to no run — a cloud, the drawing — shows its type instead. A name too long for the card is shortened; the full name, the type and the stored UTC stamp appear in the tooltip, and the properties strip at the bottom of the panel carries the detail. Data that has not been used by any run simply has no line attached — an orthophoto that no process reads stays on its own, which is the honest picture.

Relations. Input runs from a datum to the process that reads it, Output from a process to what it produced, and Mutates back from a process to a datum it rewrote in place — a classification is the one operation that does this, since the ASPRS class byte is written inside the cloud itself. DerivedFrom is a summary of the same provenance and is hidden by default to keep the diagram readable; when it is hidden the panel states how many relations are not being drawn, and the Show DerivedFrom tick box brings them back.

Moving around. The mouse wheel zooms, dragging the background pans, and Fit graph brings everything back into view. Nodes themselves are not draggable: their position is decided by the layout, not by the operator. Clicking a node selects it, and the selection is shared with the Project tab — pick a node in one, switch tab, and it is still selected in the other.

Selecting a node shows what the project actually records about it: for a point cloud, which attributes it carries (RGB, intensity, classification); for a run, its operation, when it happened and the parameters it used; for a product, its family, its status and the files it wrote. A product also states Created and, when they are not the same thing, Produced by: the first is when the node came into being, the second is the run that last rewrote its files — a raster made in the morning and recomputed in the afternoon reports both. Times are shown in your own time zone; what the project stores stays UTC and is unchanged. When a node has nothing recorded yet, the panel says so rather than inventing a value.

8.0.4 Display panel

Display — one panel that answers, for the view you are looking at, what is on screen and what could be there instead. Open it from View → Display. It carries two tabs, 3D Scene and Overview, because the two views do not show the same things: the point cloud never appears in the overview, and the surface rasters never appear in the 3D view.

It replaces the horizontal Display toolbar. Everything that toolbar carried about representation moved here; the controls that were not about representation went back to where they belong — AUTO-Z and GND-Z are in Draw → Snap, the plan-view button is View → Top (WCS), and stereo together with the display layout is in the Stereo menu, which lists every layout rather than cycling through some of them.

Three shapes of control, and each one means exactly one thing throughout the panel:

• a tick box is an object that can be on screen alongside the others — the cloud, the CAD drawing, the imagery;
• a radio button is a representation that excludes the others — the cloud is coloured one way at a time, and the overview carries one surface raster at a time;
• a slider is an option of whatever is currently chosen, and appears only when that choice makes it meaningful.

It is the only place these choices are made. The Cloud Visualization panel it replaced is gone; its colour mode, class list, point size, tint and flat colour are all here, and Cache tiles — a streaming setting, not a way of representing anything — moved to the Inspector beside the scene origin. Ctrl+Shift+L still opens this panel.

3D Scene. The theme decides how the cloud is coloured: RGB, intensity, elevation, classification, or classification tinted over the photo, plus a flat colour and a per-source tint. A theme whose attribute the cloud does not carry is disabled and says so on hover. The theme is shared by mono and stereo — there is one cloud and one colouring, and the panel states it rather than letting you discover it. Below, the options follow the theme: point size always, the ramp under elevation, tint strength under classification tint, and the class list under either classification theme, where each ASPRS class can be switched off individually. Only classes actually seen in the tiles loaded so far are listed.

Brightness and shadows. A survey taken in a courtyard, under a vault or on a north facade arrives dark, and no colour theme fixes that on its own. Two sliders sit under the theme, offered whatever the theme is: Brightness multiplies the colour and clips at white, so it is the one to reach for first; Shadows lifts the dark end without touching the highlights, which is what recovers detail in a shaded corner that brightness alone would only wash out. Both are 1.00 by default, meaning the colour exactly as the data delivers it, and a small Reset tone button appears as soon as either leaves that value. They are applied by the shader, so they are live while dragged on a cloud of any size — nothing is recomputed and nothing is re-uploaded — and they are stored with the project.

Elevation ramps. The elevation theme paints between the lowest and highest point on screen, and four ramps are offered, each drawn as a swatch beside its name because a ramp is chosen by eye. Spectrum is the original blue-green-red and stays the default, so an existing project opens looking exactly as it did. It is also the one to move away from where the picture matters: its brightness rises and then falls, so the eye reads a hard edge at the yellow band where the ground has none. Grayscale, Terrain and Cool all climb in brightness from bottom to top, which is what makes a slope read as a slope; Terrain shades it like a relief map, and Cool stays quiet enough to leave CAD linework legible over the cloud. The choice is saved in the project. A project written before this existed keeps its old grayscale setting, which becomes the Grayscale ramp.

Overview. These controls and the project tree are the two ways to set what the Zenith Overview shows; its own header names the raster on screen but no longer sets it. Imagery is a base that can be shown or hidden as a whole; the opacity of each imagery layer stays on that layer in the project tree. Beneath it, exactly one surface raster may be shown — None, or one of the DTM, DSM and nDSM this project has produced — with an opacity slider that appears once one is chosen. The same choice is on the tree's surface-model rows as an exclusive tick box; the opacity stays here. When the DTM is also the terrain the rest of the application samples for ground heights, the row says so: that is a role it carries, not a way of looking at it.

How the terrain reads. Once the project has a terrain model (or a surface raster), the Overview tab shows a section How the terrain reads with four exclusive choices for how the terrain is painted:

• Shaded relief — the default. Greys, with the light coming from the north-west at 45 degrees and the relief exaggerated twice. It is the rendering in which a terrain model is judged.
• Elevation — coloured by height, with shading.
• Slope — coloured by steepness, with a legend in degrees.
• Aspect — coloured by the direction the slope faces; flat ground, which faces nowhere, stays grey.

The choice is remembered between sessions and applies to the overview and to the raster of the mono view (10.1.1). A project that had saved another choice before Shaded relief existed opens with Shaded relief the first time.

8.0.5 UI language (3.25.2)

The UI supports English, Italian, French, German and Spanish for menus, the Scene Tree, Welcome, Project Manager, Display setup, shortcuts, toolbars, and many panel/dialog titles. Open Help → Language (available with or without a project open):

• Auto (system) — follows the Windows UI language when it matches a supported code; otherwise English.
• Or pick a language explicitly; chrome updates immediately.

CAD Draw / Modify command names stay in English (CadDll). Some long dialog body text may still appear in English until later fill-in. The choice is saved in application settings.

8.0.6 Interface size (UI scale)

Open Help → UI scale and pick a size from 75% to 200%. Like Language it works with or without a project open, so an interface too small to read can be fixed before opening anything. Menus, panels, toolbars and text change at once, and the choice is remembered. On first start the size is chosen for you from the screen (see below).

8.1 Menus

8.2 Panels and dialogs

Most auxiliary windows are independent ImGui panels: they open as floating windows that you can drag by the title bar and resize freely (see 8.4 Arranging the workspace). At a clean startup only the Welcome screen is shown; Browser, Inspector, Command Line and the built-in toolbars start closed until you open them from the menus or shortcuts.

Dialogs are summoned, not restored. Project Manager, Layer Order, Toolbars, Entity Groups, Keyboard Shortcuts and Macro Commands answer a question you asked once, so they always start closed — leaving one open when you quit does not bring it back at the next start. Working chrome does come back exactly as you left it: toolbars, Scene Tree, Zenith Overview, Command Line, Layer Manager and Entity Properties.

The amber navigation reminder above the status bar names the three mouse gestures for new operators. Closing it — either control on the band — hides it until the next installation; Help → Show navigation reminder brings it back sooner.

• Browser — project tree, navigation settings (including Kensington trackball), and quick links to Project Manager.
• Inspector — project & scene settings (navigation, magnifier, autosave, save drawing). Point-cloud color mode and classification filters live in Cloud Visualization.
• Command Line — type commands and read system feedback (see Command line).

Additional panels open from menus or shortcuts and remember their last position:

The status bar (stereo coordinatograph and cursor readout at the bottom of the main window) is not a floating panel; it is drawn as a fixed overlay on top of the viewport.

Some operations still use native Windows file dialogs (New / Open / Save / Import / Export). The license gate at first launch and a few confirmation prompts are modal dialogs.

8.3 Toolbars

Every bar is chosen in one window: View → Toolbars… (Ctrl+Shift+Alt+T). It used to be a submenu of tick boxes, which meant opening it to find out what was already on; the window shows all of them at once. They are all off by default until you enable them:

• CAD Draw — points, lines, polylines, text, and related draw commands.
• CAD Circle — circle variants (centre/radius, 2-point, 3-point, tangent…).
• CAD Arc — arc construction tools.
• CAD Dimension — linear, aligned, angular and other dimension commands.
• CAD Modify — move, copy, rotate, scale, trim, extend, and related edits.
• CAD Flow — workflow helpers (OK / Esc / Undo / Redo / Quote).
• Object Snap — running snaps, Cloud Z, AUTO-Z, GND-Z, WIRE-Z and the POLAR toggle.
• Imagery — orthophoto on/off and opacity (when imagery is present).
• Cloud plane UCS — RANSAC plane UCS helper bar.
• Custom toolbar — your own floating bar of favourite actions.

The same window's Property bars section carries the combo strips. By default the Layer combo strip can be enabled there: a compact current-layer picker alongside the floating bars. Optional combos (colour, linetype, lineweight, dimension style) can be turned on when you need them; full layer editing remains in Layer Manager (LA).

Upgrading from an older release that had a single CAD Tools bar: the first launch after update may enable all six CAD bars if you previously used the old combined toolbar.

Customize custom toolbar…, at the foot of the same window, is an editor where you add buttons from a catalog of app actions, panels and typed CAD commands. Each button has a label, a command string and an optional tooltip. Your custom toolbar is saved between sessions.

8.4 Arranging the workspace

Before you start digitizing, it is worth spending a few minutes exploring what the application offers and arranging the workspace to suit your workflow:

1. Skim the View, CAD, Draw (object snap is a submenu of it), Modify, Cloud and Stereo menus to see available commands.
2. Open CAD → Keyboard Shortcuts… (Ctrl+K) to review managed shortcuts and rebind anything you use often.
3. Open View → Toolbars… to enable CAD Draw / Circle / Arc / Dimension / Modify / Flow, Object Snap, Imagery, and the Layer combo strip in its Property bars section (plus any optional property combos you need). The Display panel is its own row in View.
4. Turn on the panels you want (Browser, Inspector, Command Line, Layer Manager, etc.) from View or their menu shortcuts, then drag each floating window to a comfortable position and size.
5. Optionally build a Custom toolbar (View → Toolbars…) with your favourite actions.

Panels and toolbars are floating only (there is no docking or tabbed dock regions): drag them by the title bar, resize from the edges, and close them with the × button. While the pointer is over a panel, toolbar, menu or dialog, the normal Windows cursor is shown and mouse clicks go to the UI — not to CAD pick/zoom/pan. Over the viewport (with no UI under the cursor), the CAD crosshair and navigation apply.

The application remembers your arrangement automatically when you exit. You can also save or reload it explicitly (see 8.5).

8.5 Saving and restoring layout

Workspace layout is stored in two places:

• settings.json — which panels and toolbars were open (uiPanels).
• imgui.ini — floating window positions and sizes.

On every normal exit the current layout is written to both files. To capture a layout mid-session without closing the application, use:

• View → Save workspace layout (Ctrl+Shift+Alt+W)

To reload the last saved layout from disk (for example after experimenting with window positions), use:

• View → Restore workspace layout (Ctrl+Shift+Alt+L)

Restore reapplies panel/toolbar visibility from settings.json and reloads window geometry from imgui.ini. Keyboard shortcut overrides and toolbar customizations also live in settings.json but are not changed by Restore unless you saved them together with the layout.

8.6 Project workspaces

Changed in 5.2. The workspace is no longer chosen from the View menu nor shown on screen: the activity is (8.8). What follows still describes the three acquisition pipelines the program recognises from the data.

A workspace says which kind of job a project is. There are three, and they are three different acquisition pipelines — different data, different tools, different questions:

What they share is the engine — point cloud, CAD, stereo, measurement, object snap, cursor elevation, cloud cleanup — which is always available whichever workspace a project uses. On top of that engine the three are genuinely different applications, not three variants of one.

One entry per pipeline

There are three pipelines, so both places that offer them — the New Project dialog and View → Workspace — list exactly three entries: Aerial Mapping, Terrestrial Mapping, Mobile Mapping. Neither opens a submenu of starting points.

Terrestrial Mapping used to ask a second question here, offering Bridges, Buildings and Monuments. It no longer does. Those three were one job seen three ways — the same oriented box with plans, elevations and sections cut through it — and differed only in which toolbars opened first, which is not a question about the survey. Projects created on any of them keep working exactly as before.

Do not confuse the workspace with a drawing template. The Drawing template field lower in the New Project dialog is a different thing: it seeds the CAD drawing with squaring and title block layers.

What a workspace changes

Which panels and toolbars are open: Scene Tree, Zenith Overview, the Workbox (Terrestrial Mapping only), the CAD draw / modify / circle / arc / dimension / flow toolbars, Object Snap, Display, Imagery, Cloud plane UCS, Layer Manager and Entity Properties. It does not change the project format, the data, the point cloud or the CAD drawing.

Your arrangement is yours

How you arrange the interface is yours, and there is one arrangement: the panels and toolbars you left open are the ones you find. Opening a project never rearranges anything — not from the workspace preset, and not from any remembered snapshot. A preset is applied only on a deliberate act: creating a project, or picking a workspace from the View menu. To put the panels back the way the workspace defines them, use View → Workspace → Apply workspace layout.

The width of the fixed CAD Tools / Scene Tree / Object Snap bands is the one thing remembered per kind of job: drag one wider on a Terrestrial Mapping job (whose scene tree runs session → families → elaborati, deeper than an aerial one) and it stays wide there without following you to the next aerial project. A band never dragged on a given workspace keeps whatever width you last set anywhere, rather than snapping to a default. Widths only — which panels are open is never decided for you.

What is never touched. Floating window positions and sizes, docking, which monitor a window is on, your stereo and display layout, your custom toolbar, and any panel you open on demand (Project Manager, Keyboard Shortcuts, Macro Commands, Browser, Inspector, command line). Those stay in settings.json and imgui.ini — the fixed CAD Tools / Scene Tree / Object Snap bands above are not floating windows, which is why their width is the one exception.

Where it is stored

The workspace and the profile are properties of the project and travel with it: both ids are written to scene_graph.json. How you personally arranged the interface stays in your own settings, so two operators can open the same project on differently arranged desks.

Projects made before this

A project created before workspaces existed has no workspace set. Nothing is applied to it and nothing about it changes; View → Workspace shows (not set) until you choose one, and the status bar names the product rather than a kind of job.

Projects created when Architecture, Infrastructure and Bridges were separate workspaces still open exactly as they did. Architecture and Bridges now open as Terrestrial Mapping, on the profile that keeps the panels they had. Infrastructure opens with no workspace set: a road survey is an aerial job or a mobile one depending on how it was flown or driven, and the file cannot say which — so the software attributes nothing and leaves the choice to you.

Projects created while this pipeline was called Built Artefacts (versions 3.86 to 3.90) open unchanged as Terrestrial Mapping, on the same starting point they were created with. Only the name changed; nothing is rewritten in the project file.

The three window buttons

Three windows are opened and closed the same way, from three buttons stacked in the top-left corner of the mono view: Scene Tree, Multiview and Zenith Overview. Each one is blue and reads Open… while its window is closed, and amber-orange and reads Close… while it is open. The same click that opened it closes it, so no window can be left on screen with no visible way back.

Multiview and the Zenith Overview are one slot, not two. Both dock into the band beside the mono view, and that band holds one panel at a time — together with the mobile station strip and the sheet index, which have always shared it. Opening one closes the other. Their two buttons are adjacent in the column for that reason: they are an alternative, not two independent switches. The Scene Tree has its own dock on the left and is unaffected.

A button appears only where its window belongs to the kind of job. Mobile Mapping shows only Multiview: that pipeline is its synchronized views plus the station strip, and neither the Scene Tree nor the Overview has anything to add to a run read that way. Multiview is offered because a roof or a road is drawn on a vehicle survey too — but it starts closed, since the band it would take is the station strip's, and opening it puts the strip away. Every window stays reachable from View and from its keyboard shortcut whatever the workspace, so nothing is ever locked away — the button says what this job normally uses, not what it is allowed to use.

Changed. Aerial Mapping now starts with the Scene Tree open and the Zenith Overview closed; before, it was the other way round. A window that costs one click to open and one to close does not have to be guessed right by a preset. Your own arrangement is still remembered per workspace (8.5), so once you have set a job up your way, that is what comes back.

Mobile views: photograph and plan / vehicle

View → Mobile views (Ctrl+Alt+3) splits the viewport in two, bound to the same station: the photograph on the left, a third of the width, and the mono view on the right. The mono view is the plan, where you draw and extract; it can also stand where the vehicle stood.

The car button on the orientation triad switches the mono view to the vehicle view and back. It is there only on a project with a mobile run, and it is lit while the vehicle view is on. The same switch is View → Vehicle view, and a keyboard shortcut can be assigned to it (Vehicle view / plan).

• The vehicle view is for looking only: nothing is drawn there. With an extraction tool armed the click is refused and the status bar says to go back to the plan.
• A click on the cloud centres the photograph and the plan on that point.
• The wheel zooms the photograph, and the vehicle view narrows with it, so the two always frame the same thing.
• Dragging with the middle or the right button, or the trackball's rotate, turns the look left and right, as the angle slider in the station controls does. The photograph turns with it to the camera that faces that way. The plan underneath does not move.
• Pressing the car button again brings back the plan exactly as you left it, or centred on the last point you clicked.

The car with a pencil, next to it, is the drawable vehicle view (View → Vehicle view, drawable). The mono view turns the way the vehicle looks, line of sight horizontal, without perspective: it is the same oblique view as the side views, so you draw, snap and extract there. So that what lies behind does not stack up to the horizon, the cloud is cut to the 30 m in front of the vehicle. It follows the run from station to station; pressing the button again brings back the plan and the whole cloud, and turning the view away ends it by itself.

Why two: the looking view is in perspective, like the photograph, and the CAD engine has no perspective projection, so a point clicked there could not be placed where it belongs. The drawable view gives up perspective to be drawable.

Changed. The separate navigation pane beside the plan is gone: the three views are now two, and the view from the vehicle is the mono view itself. The station controls are the first view of the bottom panel.

Licensing

One licence covers everything. Aerial, Terrestrial and Mobile are no longer sold separately: the kind of survey is read from the data (8.7.1), not a product you buy. Every workspace is available to every licence, whatever the licence names, and New Project and View → Workspace never list one as not included. A project opens with the workspace it was saved with.

8.7 Job types

A job type says what a job must deliver: which objects to find, which point classes matter, which deliverables (drawings, tables, a classified cloud, a report) and how each one is made, with the codes and layers of your client. Everything the program offers for a project — the steps, the graph, the chain — follows from the job type, so it is the first thing a project needs.

The catalogs shipped with the program

Every job type also has Objects to identify, Objects report and Job report.

Several job types in one project, one current

A project can hold several job types: those ticked at creation, and any you add later. One is current: the graph, the band at the bottom and the chain all work on the current one.

In the bottom panel’s Deliverable view, the What must you deliver? list has three parts. Chosen for this project lists the job types the project already holds, in normal text; choosing one switches to the project’s own copy, so classes you named and codes you changed are kept. Other job types are greyed but selectable; the tooltip says that choosing one adds it to the job types of this project. Not meant for this kind of data lists the ones the data cannot support. At the bottom, Create from the classes of the cloud makes a catalog from the classes present in the cloud, with their point counts: standard codes get their standard name and colour, the others stay to be named until you name them (open the entry, type a name, pick a colour, Save name; the legend follows at once).

Studio catalogs

A catalog of your own, with your client’s codes and names, is placed in %APPDATA%\StereoCloud3D\jobcatalogs. It survives program updates, appears in every project’s list, and carries (studio) after its name, so it can be told from a shipped catalog with the same name. A studio catalog is a good place for codes you reuse from job to job.

The project follows the program’s catalog until you change its copy

The project keeps its own copy of each job type, with where it came from. While you have not modified the copy — no class named, no customer code changed — the project follows the program: when you open it after an update that changed the shipped catalog, the copy is replaced by the new version and the status line says so. A copy you changed is never touched. Projects made before this was recorded keep what they have.

8.7.1 What the program reads from your data

The program reads facts from the data and declares them, with the reason. It does not ask you what the survey was. The facts are shown in the (data) tooltip of the graph band (8.7.4) and, at the top of the Deliverable view, as The data: followed by the survey type and, below, Classes, trajectory, photos and points per m2.

Aerial, terrestrial or mobile

The density ratio is the 90th percentile over the median of the point density in ~4 m columns of the plan. Airborne data is nearly uniform (ratio up to 4); anything scanned from the ground falls off with distance from the sensor (7 or more). The mean density alone cannot tell them apart: a drone and a vehicle can both reach 1000 points per m2. The kind of survey only guides the workspace and which job types are offered: nothing is switched off by it. If it is wrong, change it in the New Project Options.

LiDAR or photogrammetry

The graph sources tell the two clouds apart by intensity: a laser return carries a varying intensity (LiDAR cloud, laser returns); a cloud made from images has none, or a constant one (Photogrammetric cloud, from images, no intensity). The program samples a few hundred nodes of the cloud: more than two different intensity values means laser. This choice is what makes the building outlines take the right route (8.7.3).

“A surface only”

At New Project the program also reads the LAS header (the software that wrote the file and the return numbers). The note Photogrammetry: a surface only appears when a photogrammetry program is named in the header; Probably a surface only (first returns with colour) when every point is a first return and the point format has colour (aerial surveys only). It means wires and the ground under trees are probably not in the cloud. It is information, shown beside the tools that look for wires and ground; it never switches a tool off. When the headers do not say, the program says nothing; a cloud imported before this was recorded is never judged.

Colour

The data is declared without colour when the cloud has no colour channel or the colour fields exist but every sampled point is black. For a mobile run whose photographs are in the project, the Deliverable view then offers Colour the cloud from the photos (the same command as Cloud → Colour the cloud from the run). It paints every point with the colour of the photographs the vehicle took and rewrites the cloud in place; it can take several minutes.

Noise on flat surfaces

At import, when the octree is finished, the program picks 12 windows of 40 m spread over the whole survey, reads every point in them, and measures the distance of the points from the best plane in flat 0.5 m cells. The figure shown is the median of the windows with the range from the best to the worst: Noise on flat surfaces: 13.3 mm (median of 9 zones; from 5.0 to 24.2 mm). It is an upper limit of the instrument noise — real roughness of the ground and drift where the same place is seen more than once add to it — and it cannot be compared with a manufacturer’s stated accuracy. A cloud imported before this existed says not measured; a cloud with no flat surface, or too sparse, says not measurable. A failure never stops the import.

8.7.2 The Automation menu

The Automation menu, between Cloud and Stereo, collects the commands that run for a long time and produce results:

• What can I do with my data… — the graph (8.7.3).
• Classification — the same cloud classification passes as in the Cloud menu (Classify cloud (all passes) first, then the individual passes).
• Buildings — Building outlines preview…, Regenerate building hypotheses…, Building outlines check….

When the project has a job type, a command that is a step of that job — chosen from a menu, with its keyboard shortcut or typed in the command line — does not start by itself. It opens the graph on its product, and the status line says This step runs from the graph, so the sequence stays consistent: choose it there. This keeps the sequence of steps coherent: you cannot redo the classification from a menu and leave contour lines made from the old classes. The buttons inside a working panel (bridge, track, reconstruction) are already the step in progress and start at once.

8.7.3 What can I do with my data

Changed in 5.2. The Automation menu is replaced by the activity menu (8.8.3): this window opens from Full graph in the Project map (8.8.1).

Automation → What can I do with my data… opens a window with three columns: Sources (what is in the project), Products and Next steps / Final products, joined by lines.

• Sources. Read when the window opens, with a wait cover while it reads: LiDAR or photogrammetric cloud, classes in the cloud, orthophoto, CAD drawing, trajectory and photos. Each box says what it holds (classified / not classified, N lines in the drawing, N terrain rasters, laser returns or from images) or not in the project. Green is in the project, grey is missing.
• Tabs. One tab per job type, read from your data. The first, This project: …, is the project’s current job; the others are Buildings and terrain, Railway, Power line and Engineering structures. Their state is read from the same data, so you can see what each job would need. Starting a step from another tab makes the project work as that job (The project now works as: …); a tab that is not the project’s says This is not the job of the project yet and offers Work on … in this project.
• Boxes. Each box is a product, with its state under the name: teal done (click to redo), orange to redo, grey waiting (the tooltip says for what). A delivered table whose objects are no longer in the drawing (the mast inventory after the masts on PALI_TE were deleted) is not done any more: it goes back to waiting for them. Click a box to see Comes from and the methods that make it. Click a delivered file’s box to open it.
• The play button on a box makes that product; once it is done the same button is a circular arrow and redoes it. When several processes can make it, a list opens beside it, with the recommended one first (Classify the cloud, Automatic outlines (local)). A grey entry says why it cannot run. A box with a bright green border is a step you can take right now.
• Do everything up to here — on a deliverable that is not yet delivered, a second button (two triangles) runs in order the automatic steps still missing before it (8.7.5).
• A step needs what is upstream: with the earlier product missing, or marked to redo, the box is off. Redo that one first.

The recommended method

The recommended method is the one the program takes with these data. Building outlines are the example: on a LiDAR cloud they come from the cloud, whose edges are sharp, and the orthophoto is not used. On a photogrammetric cloud with an orthophoto the roof edges are taken from the orthophoto and the heights from the cloud. On a photogrammetric cloud without an orthophoto they come from the cloud and can be less sharp; with an orthophoto in the project the result is better. The Process view (below) states which of these routes will be taken.

Redoing a step

Repeating a step makes what came after it out of date. The rule, applied when the step really starts (cancelling changes nothing):

• Computed drawings are replaced — a drawing the program calculated (contour lines, outlines it generated) loses its previous version when the step is made again, so old and new do not pile up. Objects you drew or corrected by hand are not part of this.
• Drawings made by hand are hidden — the objects of the job (conductors, buildings you confirmed) are not deleted: their layers are switched off until their step runs again.
• Delivered files stay and are marked to redo — tables and exported clouds are kept, shown in orange, and go back to good when their step has run again.

The Process view lists, before you press Run, exactly what will happen to each product that comes after (will be deleted, will be hidden, to redo). When a classification, terrain, contour or outline run is started by another route, the program shows a message listing the products that are now out of date, with a button to open the graph. Nothing is deleted without being listed.

8.7.4 The graph band and the Process view

Changed in 5.2. The band at the top of the bottom panel is replaced by the activity bar under the menus (8.8.2). The Process view described below stays: it is where a step clicked in the activity bar opens.

The top of the bottom panel (see 9.2.2) holds a band that keeps “where we are” in view, whatever view is in front. It appears when the project has a job type.

• Closed (one line): the arrow, Full graph (opens the window of 8.7.3), the job type, (data) (hover for the description of the data), N of M done, N to redo in orange, and next: … with a green Run: … button. The button opens the step in the Process view; when several methods are possible it takes the recommended one. When nothing can run it says nothing to run now: see the full graph, or everything done.
• Open (click the arrow): the graph in miniature — the same boxes in colour, with the cloud classes grouped in one box (Cloud classes; click it to choose the classification in force, the same control as in the Display panel). Click a box to make or redo it; a delivered box opens its file.

Open or closed is remembered. Closed by hand, the panel stays closed (9.2.2).

The Process view

The graph chooses; the Process view of the bottom panel shows and starts. It comes to the front when you choose a step, and shows:

• The product and the process, and Open the graph.
• When it starts, what came after it: … (in orange, with the fate of each product), or Nothing made after this step yet: running it changes nothing else.
• For building outlines, a line saying which route the program will take with these data.
• The options of the step. Cloud classification and contour lines show their options right here, in columns, instead of a dialog. Other steps have a Run button, or Options for those that open their own panel. Run is off with the reason as tooltip when the data cannot support it, and the line Waiting for: … names the missing earlier product.
• A product already made is not made again with one click. When the step is already done the button reads Redo it and asks on the line below, with the date: Terrain model (DTM) is already done (Delivered …). Running it again replaces it. — Yes, run it again or Keep it. A click by mistake no longer repeats a long computation.
• The result. After Run, the first lines the step wrote appear under the button: the status line of the command, errors in red, warnings in orange, the rest in normal colour. Periodic diagnostic lines are left out. If a step fails, the cause is here, not only in the log.

8.7.5 Run the chain

Many deliverables need several steps in a fixed order — classify, find the wires, find the supports, write the tables. The chain is that list for the current job, one step per method, in the order of their dependencies. Run the chain starts the automatic steps one after the other, from the first one that is not green, so you do not have to press Search after every step.

• Drafts are confirmed by the chain only when green — that is, when they meet the measured criteria. Yellow and red ones wait for you. If the catalog gives no tolerance for the conductors, green cannot be measured and the chain says so.
• Manual steps. A step the chain does not start (it needs a gesture of yours, or it has no automatic outcome) is shown as manual and is yours, with its own button. It does not stop the automatic steps that do not depend on it: the chain goes on, and at the end says which step is left to you (Your turn: step N, …). Optional steps are marked (optional): the chain is complete without them.
• The line beside the button says what to do now: where the chain starts, the yellow to resolve (with the number to check), the red with its reason, the manual step that is yours, or Chain complete: every step is green. Run the chain is on only when there is an automatic step to start. Stop stops it after the current step.
• Every step done saves the drawing. When a row becomes done (delivered or confirmed), the drawing is saved before the next frame, with a backup copy in cad/backups (see 6.5). The status line says Drawing saved: … done, or, if it could not be saved, tells you to save with Ctrl+S. A “No” at the close-and-save question can therefore no longer lose a step that has already delivered its tables.
• Cloud classification runs inside the chain without its dialog, with the passes you chose last time. The chain waits until the class counts have been read again. For jobs that have no classification (the railway) it does not classify.

You start the chain from the graph with the two-triangle button on a deliverable (Do everything up to here). While it works, or while it waits for a confirmation, the Deliverable view shows its row of steps: for each step its name, one line of state and its buttons, with the products (with counts and N to check) under it, and Confirm / Discard under the drafts. The row scrolls to the step the line is talking about.

8.7.6 Delivered files and tables

• Open opens the delivered file with its own program (or the folder, for a set of files). In the graph, clicking the box of a delivered product does the same.
• View, beside it, shows a table in the program’s own viewer: a window with a fixed header, scrolling, resizable columns and the number of rows (the first 50,000 are shown, with a notice, the file holds the rest). From it, Open with its program and Show in folder. A delivery with several tables (the railway) lists them to choose from.
• The support, span and warning tables of the power line, and the cross sections, are written as CSV and also as an Excel (.xlsx) file with the same name next to it: real numeric cells, bold frozen header. The CSV is unchanged. If the .xlsx cannot be written the status line and the log say so.
• The classified cloud is exported without a dialog inside the chain, to exports/nuvola_classificata, only when the original file can be found (among those remembered for the project, or in the folder the survey data came from, with the same name and the same number of points). Without the original the step is red with the reason, and its button opens the export dialog once.

8.7.7 The Deliverable view

Changed in 5.2. Inside an activity the Deliverable view does not open by itself any more, not even while a draft waits: drafts are confirmed from the activity bar (8.8.2). It opens from the Deliverables button of the Project map. In mobile mapping it stays as it was.

The Deliverable button of the bottom panel is now needed in three cases only:

• Choosing the job. A project with no job chosen asks What must you deliver?, with the trades proposed from what the data contains (classes already named, the signature of a trade, the kind of survey), and the same list as in 8.7. Nothing is put in force by guess: you choose.
• The chain at work or waiting for a confirmation (8.7.5).
• Mobile mapping, where it stays as it was: the delivery rows with their state and their buttons.

At other times the bottom panel is the graph band alone, and it does not open a view on its own. The things that used to be in the view are elsewhere: Open on a delivered product is the click on its box; the classification version is the Cloud classes box; the description of the data is the (data) tooltip; Add data is File → Add data… (6.1).

Several states, one place. The graph is where steps are started. The Deliverable view shows; the graph starts. A delivery row’s In the graph button opens the graph on that deliverable.

8.8 Activities: one job on screen at a time (5.2)

From 5.2 the interface is organised around activities. An activity is a result you recognise — Stereo plotting, Auto buildings 3D, Terrain, Railway — with its own screen: its steps, its menu, its panels and its views. Only one activity is on screen at a time. The data are shared: the cloud, the drawing and every product are the same whichever activity you are in; an activity changes the tools, never the data.

There are two places only, and each answers one question:

The activities are:

Workspace and profile are no longer on screen. The kind of survey (aerial, from the ground, mobile) is a fact of the data, read when the project is created; it decides what is possible, but you do not choose it as a mode. The status bar and the window title name the activity, not the workspace, and View → Workspace no longer offers Aerial / Terrestrial / Mobile Mapping. A wrong verdict is corrected in the Options of the New Project dialog.

8.8.1 The Project map

The Project map opens every time a project is opened, and from the Project map button at the left of the activity bar. It takes the whole window below the menus.

• Your data (left): LiDAR or photogrammetric cloud, classes in the cloud, orthophoto, CAD drawing, trajectory and photos. Green is in the project, a grey frame is missing.
• What you can do with these data (centre): one card per possible activity, with a state read from the project — IN PROGRESS (the activity you were in), DONE, PARTIAL, READY — and a line such as 1 of 3 steps done, 2 to check, 1 to redo. Open enters the activity; when the project does not have its job type yet the card says not in this project yet: Open adds it, and Open adds it.
• Try on an area (Auto buildings 3D): place a square on the cloud and the building outlines of that area appear in a preview window. Nothing is written until you confirm, and you can cancel at any time. It is the quickest way to find out what the automation gives on your data.
• Not possible with these data (bottom, closed): each activity that cannot run, with the reason in one line — needs a survey of type: terrestrial, mobile, or no points of its classes (10) in the classified cloud. The second kind can still be opened with Open anyway.
• Resume: … (top right, or Esc) closes the map and takes you back to the activity. Deliverables opens the Deliverable view (8.7.7) and Full graph the graph of every job catalog (8.7.3).

If the map opens while the program is still counting the classes of the cloud, it is rebuilt by itself when the count arrives, so activities whose classes are missing move to the bottom list.

8.8.2 The activity bar

One row under the menus, always in the same place:

• Project map — back to the map.
• project > activity — where you are. The name of the activity is the orange button: click it to change activity (the list shows the activities the project’s job types carry; the others are in the map).
• The steps of the activity in order, each with a numbered disc: blue to do, yellow / amber to check, green with a tick done, orange to redo (made from data that changed since), grey waiting for an earlier step. Hover a step for its state; click it to open the step in the Process view of the bottom panel. With several methods it asks which. A step that is a place to go rather than a process to run — Sheets, the drawings made in the Workbox, the bridge volume, the tunnel panel — takes you straight to its panel instead.
• For Railway and Power line the steps are those of the chain (8.7.5): one per method, red when blocked, and an orange disc with a ring when a draft waits — click it for Confirm or Discard. Run the chain sits at the end of the bar when there are automatic steps to run.

The prerequisites of a step are inside the step: the classification is never offered as “the next thing to do” unless a step of the activity you are in needs it.

8.8.3 The activity menu

With a project open, the menus Terrain and Automation are replaced by one menu named after the activity — Auto buildings 3D, Terrain, Railway… — holding the commands of that activity, with the same shortcuts as before, and Project map at the bottom. From Cloud the trade commands moved to their activity (export classified cloud, railway, conductors); the general ones stay (representation, measuring, colours, cleanup, performance, Draw overhead line). Without a project the menus are as before.

8.8.4 Changing activity: what happens to the screen

• The windows of the activity you leave close. Building review, Multiview drawing, contour lines, volumes, sections, railway, power line review, measuring, photo pairing… close as their own Close button would, and a tool left armed (measure, volume area) is disarmed. The status line says how many windows were closed. Calculations running with their own progress go on; windows holding unsaved work (vehicle masks) are left open.
• Each activity remembers its own arrangement of panels and bars, and finds it again when you come back. The first time, it takes the arrangement the job type had.
• Multiview exists in Mono plotting, Auto buildings 3D and Roads (its Road tab); the Buildings list only in Auto buildings 3D. Entering another activity closes them.
• The bottom panel shows the step you are on, or the views the activity uses (vertical profile, breaklines, volume profiles). The Deliverable view no longer opens by itself inside an activity: it is the Deliverables button of the map.
• The CAD drawing stays on the cloud when panels open or close and the view changes size.

8.8.5 The i button of the view

The yellow box with the zoom (px/m, points shown, octree nodes) and the Safe display mode notice are behind a small i button in the top right corner of the view. Click it to show or hide them; the choice is remembered. The button turns orange when the safe display mode is on, and its tooltip says what that means. Nothing is drawn over the Project map.

9. Navigating & snapping

9.1 View navigation

• Top (T) — orthographic top view.
• Fit Extents (F) — frame the whole scene.
• Unlock input (Ctrl+Shift+R) — clears stuck pan/orbit sessions, closes escapable modals and reclaims keyboard focus without changing the current view (see 9.7).
• Restore view and alignment — recovers input, then restores zenith WCS and re-syncs cloud, CAD and camera (see 9.7).
• Right mouse button (RMB) drag — rotates the model. Hold the right button and drag to orbit the view around the cloud point at the centre of the window, resampled as you drag. This is the primary rotation gesture, the one most 3D software uses. A right click without dragging still opens the framing/context menu (see framing commands). It works inside a draw tool too: the drag rotates without closing the polyline, and a right click without dragging still closes it.
• Middle mouse button (MMB) drag — pan the view when idle (cloud and CAD move together). Shift+MMB drag also rotates, kept as a fallback. While a draw tool is active, MMB drag orbits instead (see 9.2).
• Mouse wheel — live zoom toward the cursor. The wheel speed is adjustable in the Inspector (default 2.0). The cloud uses a temporary lower resolution while the wheel is moving; full detail returns shortly after you stop.
• Show cloud (Ctrl+Shift+C), Show CAD (Ctrl+Shift+D) and Show imagery (Ctrl+Shift+I) switch a whole kind of content on and off. Their tick boxes are the Clouds, CAD and Imagery rows of the project tree — the rows that name the content — and the shortcuts do the same thing. Cloud on mono view (Ctrl+Alt+C) hides the cloud on the mono UI only while stereo keeps it; it is a property of one view, so it lives in Display → View.

Framing commands (right-click menu)

Right-click on empty space in the viewport opens the CAD context menu. Besides Zoom Extents it offers the framing commands below. All of them act on the georeferenced mono view; when that view is not the active one the command says so in the log instead of doing nothing.

• Zoom Window — pick two opposite corners in the viewport and the view frames that rectangle. Drag from one corner to the other, or click once on each corner; both work. The rectangle is widened to the shape of the viewport, never cropped, so everything you enclosed stays visible. A rectangle smaller than a few pixels is refused rather than throwing you into an unreadable scale. Esc or a right-click cancels while the corners are being picked, and so does starting any other command.
• Zoom Previous — returns to the framing you had before the last zoom or pan. The application records one entry per zoom-extents, per zoom-window, per wheel burst and per pan drag, so repeated calls walk back through the framings you actually recognise instead of one wheel notch at a time. Orbit is deliberately not recorded: this command restores the framing, not the orientation.
• Zoom In / Zoom Out — halve or double the framed area around the centre of the view. Same effect as the wheel, in one fixed step, for when you want a repeatable amount rather than a gesture.
• Zoom Selection — frames the selected entity, with a margin so it does not sit against the edges. The entity's own vertices are projected through the same projection that draws the scene, so it works in zenith and in oblique alike, and it reaches a selection that is currently off screen. The menu entry is greyed out when nothing is selected. With several entities selected it frames the one the CAD engine reports as current, not their combined extents.
• Pan — while it is on, dragging with the left button moves the view, exactly as a middle-drag does. It stays on until Esc, a right-click, or any other command ends it; the status bar shows Pan for as long as it is armed. The middle button keeps working throughout, so this command is a convenience, not the only way to pan.

Each one also answers to its typed command — ZW/ZOOMW, ZP/ZOOMP, ZI/ZOOMI, ZO/ZOOMO, ZS/ZOOMSEL, P/PAN — and can be given a key of your own in Keyboard Shortcuts.

Navigation reminder band

The first time a project is opened, a yellow band appears directly above the status bar listing the mouse gestures: wheel to zoom, middle button held to pan, and the right button held to rotate the model. It is there because the rotation gesture is the one nobody guesses.

The band takes its own strip and never covers the 3D view. Both controls do the same thing — the x and the Do not show again box hide it until the next installation, and installing a newer version shows it once more, because that is when the gestures are worth naming again. Help → Show navigation reminder brings it back at any time.

9.2 Orbit on three axes (cloud pivot)

StereoCloud3D lets you rotate the view around any point on the cloud — essential for inspecting façades, bridge soffits, dam faces and other oblique work before you lock in a Stereo UCS wall view. Orbit uses a pivot on the cloud surface and supports yaw, pitch and roll (three independent axes).

The pivot takes itself

• Point at the place you want to turn around and start dragging with the right button. The pivot is sampled on the cloud under the cursor at the instant the drag begins, and the view turns about that point until you release.
• Every drag takes its own pivot, so you change the centre of rotation simply by starting the next drag somewhere else. There is nothing to set beforehand and nothing to clear afterwards.
• If there is no cloud under the cursor the pivot falls back to the view centre, as it always did — the rotation still works, it just turns around the middle of the screen.
• With vertical exaggeration on, the pivot lands on the point you pointed at, and the view turns about it in both the cloud and the CAD.

Rotate around the pivot

• Shift + MMB drag — yaw (horizontal) and pitch (vertical) around the pivot. Cloud and CAD rotate together.
• Ctrl + Shift + MMB drag — roll (twist) around the view axis through the pivot.
• Alt + right-button drag — turns the model around the vertical axis (world Z) through the pivot, like a turntable: the tilt you have chosen stays exactly as it is and only the direction you are looking from changes. The plain right-button drag turns around the axis that is up on the screen, so in an oblique view it brings the horizon with it; hold Alt when you want to bring the far side of a building towards you without losing the inclination. Only the horizontal movement counts — the vertical is ignored, because Z is a single axis. It works inside a draw tool too, without closing the command, exactly like the plain right-button drag below.
• Right-button drag while a draw tool is active — rotates, taking its pivot under the cursor as usual, without closing the command: the polyline stays open and the vertex you are about to place stays where it is while the view turns around it. What tells the two apart is movement: a right click without dragging still belongs to the command and closes the entity (see the note below), handed to it when you release the button. With Shift held the right button never reaches the command at all — a Shift + right click does nothing — for when you want to be sure a hesitant hand does not close the line.
• While a draw tool is active, MMB drag orbits instead of panning; the rubber band and command stay live when you release the button.
• Ctrl + LMB drag while drawing also orbits (StereoCAD-style) without placing a vertex.
• While a pivot session is active, an RGB axis gizmo (X/Y/Z) is drawn at the pivot to show the current orientation frame.

Performance while orbiting

During Shift+MMB drag the cloud temporarily uses a coarser level-of-detail so rotation stays fluid on large datasets. When you release the middle button, full resolution is restored automatically (same behaviour as pan/zoom). When you zoom far away from the subject, the cloud deliberately drops detail for navigation; close-up restitution keeps the high-detail LOD and transitions progressively as you move away from the subject.

End the session

• T (Top / WCS) restores plan framing when you return to cartographic work.
• Esc cancels the current CAD command; it has no effect on the orbit, which owns no state between one drag and the next.

Older versions used the G key to place the pivot before rotating. It is no longer needed — the drag does it — and is no longer listed among the shortcuts. The key still answers for anyone who has it in their fingers.

9.2.1 Multiview — the drawing workbench

Changed in 5.2. The two views follow the cursor: when it stops on the view for about a third of a second (and has moved at least 12 px), they are taken again around the point under it, before the first vertex too (as views only: there is no vertex to correct yet). A new vertex catches them as before, and the height and offset fields stay the vertex’s. Opening Multiview turns AUTO-Z on; a height locked by hand (Locked Z, wheel) still suspends it. Multiview belongs to Mono plotting, Auto buildings 3D and Roads (8.8.4).

The working view is nearly always looking straight down, and from there elevation cannot be judged: a vertex thirty centimetres above the roof and a vertex sitting on the roof occupy the same pixel. View → Multiview opens the panel that answers that question, and around that answer it gathers the drawing work that needs the answer. It draws nothing on its own initiative and changes nothing by itself: everything it writes, you asked for.

Six tabs, and what they share

The panel is divided into tabs, one per kind of object you deliver: Roof (pitches, eave, foot, walls — all parts of one object, so one tab), Road, Polyline, Spline, Ellipse and Rectangle. Modes inside a tab are ways of drawing that object, never a different object.

The point and the two views live below the tabs, not inside one of them. They are the panel's answer, not the roof's: a point is acquired by drawing anything, so the coordinates, the elevation field, the offset, the Options and the two rendered views stay on screen whichever tab you are on. So does the loupe checkbox, which is needed while you aim.

Leaving a tab stops the command that tab started. A tab is a command: leaving one with its command still armed would have you drawing one thing while looking at another. Every drawing command is stopped on a tab change — the application's own (pitch, eave, road) and the engine's, which are reset the way Esc resets them.

Two conditions before it will work

Multiview reads elevation off the rendered frame — depth taken from the image and converted back into coordinates. Two things make that reading wrong without looking wrong, so the panel refuses to work while either holds, showing a translucent cover over its own content with the reason written on it:

• The view must be nadir (looking straight down, within 1°). Off nadir, the few pixels a sample reads can land on a surface that is not the one under the cursor; the cover tells you how many degrees the view is tilted. The title bar stays clear of the cover, so you can always close the panel.
• Vertical exaggeration must be 1×. This one the application enforces for you: opening Multiview holds exaggeration at 1.0× for as long as it stays open, and an amber line in the panel says so and names the value it will restore when you close it. Your saved setting is never changed.

While the cover is up no capture is taken, so a tilted view also costs nothing.

The two views

The panel shows the last acquired point — the point a draw command actually consumed, object snap included — from two real photographic views, both 45° above the horizon, rendered by the same engine as the main view rather than calculated from a list of points. Each carries the point theme, the point size, the LOD and the streaming of the real view, and shows the cloud's true colour where available. The corner of each view says where it looks from, in degrees from north: from 90 deg is a view taken from the east, looking west.

• The bearings are chosen by looking at the cloud. Thirty-six candidate bearings are scored against the point cloud: for each one, the line of sight from the point to the eye is tested, and the points that lie within a thin cylinder around it are counted as obstacles. The two clearest win, kept between 90° and 120° apart and never opposite — two views ten degrees from each other say nothing more than one, and two opposite views look at the same vertical plane from both sides and leave the other direction free. When nothing is in the way — an open roof, the normal case — the panel keeps your own bearing and your bearing plus 90°, and while nothing blocks the two views already in use they stay as they are, so the views do not wander from one stop of the cursor to the next. The mechanism only engages when there is a real obstacle to avoid. The result, with the obstacle counts that justified it, is printed under the option so you can judge the choice instead of trusting it.
• The point stays in the centre, the images move. With the cursor on the mono view, the reference of the panel is the point under the cursor: it stays at the centre of both views and the two images slide under it as you move, every frame. When the cursor stops, the two views are captured again around it. The coordinates, the Z field and the elevation line are those of that point, and the vertex you have in hand stays visible as a hollow circle, with a line under the coordinates saying so. Move onto the panel and the reference goes back to the vertex, which is where you correct it; if the vertex is outside the images taken around the cursor, the views are captured again on it.
• The 3 cm magenta sphere — the acquired point is drawn as a sphere 3 cm across, to scale, in magenta: no cloud produces that colour, so it never blends into the points. It is a ruler, not a pointer. If the cloud surface covers it, the point is on the surface; if it floats two diameters above it, that is six centimetres. A thin ring with four ticks sits outside the disc — that is how you find it and grab it, and it never falsifies the measurement.
• The elevation line — a magenta line crosses each view at the point's height, labelled with the elevation in project coordinates. It is how you see at a glance whether the point sits on the surface or floats above it.
• The pitch outlines are drawn on the views. Every pitch outline near the point is projected into each view through the very matrix that produced that frame, so the outlines fall on the points that surround them. The part of an outline that is hidden behind the roof in that view is drawn faint rather than removed: it says the edge is there and where, without pretending it is in front. A dot marks each vertex, because on a long edge a vertex is otherwise indistinguishable from the segment. The outlines are re-read from the drawing every frame, so a vertex you have just moved appears immediately.
• Correcting the point — drag the sphere in either view and the vertex follows it, resolved from the two directions that view actually shows (its own bearing and the vertical) together. The drawing changes when you release, so one correction is one undo step. The gesture always answers: the sphere goes where you put it. What can fail is writing it, because a vertex a running command still holds is not an entity yet — except while an eave outline or a pitch face is being drawn, where the vertex lives in the application itself and always accepts the correction. When writing fails the panel says so in one line, and finishing the polyline makes the correction land.
• Zoom & pan — the mouse wheel zooms into the image already captured (1× to 24×) and keeps the reference point in the centre; both views zoom together, since they show the same point at the same resolution. The middle mouse button drags the images by hand, and from then on they no longer recentre by themselves. Double-click the middle button to go back to the centred view.
• Orienting the set — drag with the left button anywhere except the sphere to turn both views together, keeping their separation; useful to line one of them up with a wall. Turning them by hand switches off the automatic choice for that drag.

A capture is two off-screen renders with the tile selection redone for each pose, and on an area not already in cache it can take a moment; the panel covers itself with a progress message while it works, and does not re-capture on every small adjustment — only once the point, the radius or the bearing has settled. When the capture finishes, the tile selection is put back to the operator's own camera: without that the main view would keep drawing the handful of tiles chosen for a five-metre close-up, which is exactly what used to make the cloud go coarse after every capture.

The point, in numbers

• X, Y, Z of the reference point, in project world coordinates: the point under the cursor while the cursor is on the mono view, the last acquired vertex when it is on the panel.
• Z — the elevation as an editable field. Type the value you want and the vertex takes it: on a roof edge with a known height, typing is exact where dragging is a negotiation.
• Offset — a signed displacement in metres applied to the elevation, for when you know the correction ("this is 40 cm too low") rather than the answer.

The cursor loupe

Vertical loupe on the cursor — a checkbox in the panel, View → Cursor loupe, or Ctrl+Alt+L. The three are one command, not three copies of it: whichever you use, the checkbox and the menu tick follow, and the key can be reassigned from the keyboard shortcuts dialog like any other. It opens a small round window beside the cursor showing a vertical section of the cloud through the cursor: the horizontal axis runs along the section, the vertical axis is elevation, and a slab of ±35 cm around the cursor is kept, so you see the profile of the surface the point is about to land on. A crosshair with a filled dot marks the exact point — the arms say where it is, the dot says which it is — and the elevation is written under the disc.

It is a real render of the cloud, not a drawing, and it follows the cursor. It also updates while you nudge the elevation with Shift+wheel (fine) or Ctrl+wheel (coarse), so you can watch the point rise onto the surface instead of guessing. It sits beside the cursor, never on top of it: covering the point you are aiming at would make it useless exactly when it is needed.

Roof tab — drawing the roof

The commands for a roof are here, next to the views that check them.

• Draw pitch face / Stop drawing pitches — starts (and stops) the pitch face command. While it runs the button turns amber. Each closed pitch immediately starts a new one, so a whole roof is drawn without touching the panel again.
• Filled (solid face) — the pitch is delivered as a filled solid instead of a plain outline, with a dark border added on top so its shape reads against a dull fill colour.
• Semi-transparent 3D faces — dithers every filled face so the cloud and anything behind it stay visible through it.
• Free vertices onto the pitch plane — described in 13.7; from the fourth vertex on, a vertex you did not fix yourself takes its elevation from the plane of the pitch rather than from the cloud sample under the cursor.
• Eave layer and Pitch layer — the layers assigned to those entities (defaults GRONDA and FALDA), not whichever layer happens to be current. Both persistent.

Polyline, Spline, Ellipse, Rectangle — the CAD's own commands

These four tabs contain no geometry of their own, on purpose. The engine already draws a polyline, a spline, an ellipse and a rectangle, with its own snap, grips, rubber band, polar tracking and keyboard input; re-doing them in the application would mean losing all of that and rewriting what already works. So each tab does two things: it sets the layer and it starts the native command. Everything after that is the CAD's.

• Layer — a field per tab, persistent, created if it does not exist. Changing it while a command is running re-applies it, so the change takes effect instead of being silently refused.
• Polyline — a Closed checkbox, not a choice between two polylines: a polyline is open, and you close it. Ticking it while drawing restarts the command closed.
• Spline — its own tab, because in the engine it is its own entity with its own command: it is not a mode of the polyline, and you cannot switch between them half way through.
• Ellipse — the engine exposes a single ellipse command, so there is one mode here and none invented on top of it.
• Rectangle — two modes, and they are the two the engine has: opposite corners, or three points for a rotated rectangle. The mode is named in the button, because it is persistent and a button that always reads the same lets you draw the wrong thing for days without noticing.
• The button says what state you are in — amber and Stop drawing… while a command of that tab is running, and pressing it stops the command. Whether one is running is asked of the engine: a polyline tool re-arms itself after each entity, so the application's own bookkeeping cannot answer it.

Our own mark, on top of the engine's rubber band. The picked points carry the same sign as pitch, eave and road — light blue, a filled dot per vertex — so the panel speaks one language. Nothing is re-drawn: every time the engine consumes a point it says so, and that point is echoed. For polyline and spline the chain between the points and the live segment to the cursor are drawn too; for the rotated rectangle and the ellipse only the points are, because their shape depends on the order in which the engine asks for its data and a guessed preview would argue with the real one.

The right button closes the current entity and starts another of the same kind — the tool re-arms itself — so the echoed mark clears and starts again with it. To turn the model without closing what you are drawing, drag with the right button: only a right click that does not move closes the entity (see Orbit on three axes).

Roof tab — buildings

The panel reports how many buildings the pitches form and how many are still in draft, and carries the settings that turn a draft into drawn geometry (13.8 and 13.9).

• Eave overhang (m) — how far the foot is set in from the outline, and how wide the eave soffit is. Persistent, default 0.50 m.
• Foot layer and Wall layer — defaults PIEDE and PARETI.
• Confirm building N — writes the eave outline, the foot and the walls for the building you have selected, and marks it confirmed. There is deliberately no "confirm all": a button that writes geometry for buildings you are not looking at is a button that gets pressed by mistake.
• Delete pitch — press it, then click the pitch in the view: that pitch is deleted. With pitches one inside the other (a dormer on a roof) the smallest one under the cursor is taken. A click outside every pitch leaves the command waiting; pressing the button again cancels it. It is off while a pitch or an eave is being drawn, because there a click is a vertex, and it switches itself off when Multiview is closed. It is the same deletion as the right-button menu on a handle.
• Delete building — deletes the selected building: its pitches, and with them the eave, the foot and the walls, which the numbering removes by itself once they have no pitch left. It asks first, on the same line: Delete building N (n pitches, with eave, foot and walls)? — Yes, delete it or Keep it.
• Both can be taken back with Undo geometry edit.

Roof tab — correcting a building already drawn

Click the numbered circle of a building and it enters edit: its pitches are taken out of the drawing and held by the application, where they behave like a pitch you are still drawing — an elastic outline, with no engine in between. They return to the drawing when you leave or confirm, so a correction in progress never leaves half-written geometry behind and the fills are not rebuilt at every gesture.

Leaving loses nothing. Leave edit puts the pitches back as you left them; Confirm does the same and regenerates the eave, the foot and the walls. Saving or closing the project hands the session over first, so a building cannot disappear from the drawing because a panel was open. The panel says which building is in edit, with the way out, Undo and Delete pitch next to it; to delete the whole building, leave edit and use Delete building.

The tools are where your hand is. With a building in edit, holding the right button still on a handle opens the editing menu — delete vertex, split edge, delete pitch, undo. The right button dragged still orbits, so the menu opens only when the pointer has not moved.

Esc during a handle drag abandons it and puts the handle back where it was, instead of having to set it down somewhere wrong and move it back by eye. Everything written here can be taken back with Undo geometry edit.

The vertical profile

In the bottom panel, under the view, the vertical profile shows the cloud cut vertically through the work in progress. It is drawn as a profile line, the way a survey section is read, and not as a stretched picture of the slice.

• What it cuts. The section runs through the plan centroid of the building being worked on and is twice its largest dimension long, so the surroundings are visible and not only the house. While you review vertices one by one it follows the picked vertex instead: correcting an eave while looking at the ridge helps nobody. Direction and length stay the building's, so two sections of the same house remain comparable.
• Where it points. With pitches present the profile orients itself across them — along the slope, which is where the shape of a roof shows. It follows the pitch you are still drawing as well, which is when the profile is worth most: the slope is being decided. Drag the bearing to turn it.
• Cross section. A second line cuts perpendicular to the first — to the current segment while you are drawing — so both directions are read at once.
• Slice sets how thick a slab of cloud is collapsed into the profile.

When the profile cannot follow the pitches, the view says why — how many pitches it found on the layer and what was missing — instead of quietly falling back to an arbitrary axis.

The panel can be put away when the page is small and that space is worth more to the drawing — see The bottom panel.

The profile under the cursor (stereo restitution)

Changed in 5.2. Not only in stereo restitution: in every activity, without a Multiview work area, the vertical profile follows the cursor. With Multiview open and its work area placed, it is the profile of the area, as before.

In direct stereo restitution, with Multiview closed, the vertical profile follows the 3D point of the cursor — the same point the status bar reads — and shows two cuts side by side, crossing at the cursor:

• Longitudinal profile (left, yellow trace in the view) — along the horizontal of the screen, turned by the Rotation.
• Cross section (right, green trace) — at 90° to it. While you draw a line it is perpendicular to the segment you are pulling instead.

Each graph is a vertical cut of the cloud: distance along the line across, absolute elevation up. The line is the highest point of the slice in each short stretch, so it shows the top of what is there — grass, shrubs, kerbs — not only the ground. Both traces in the view mark with 0 the end where their graph starts.

• Width — one slider for both profiles, 0.5 to 5 m, so the two graphs have the same horizontal scale. It is remembered.
• Slice — the total thickness of the slab of cloud taken: the points within half of it on each side of the line make the profile.
• Rotation — move the slider (Ctrl+click to type the angle), or drag left/right inside the left graph (the pointer becomes a double arrow); the cross turns in the view while you drag. 0 puts it back on the screen horizontal. It is remembered.
• The cursor — a green vertical line with its elevation. When the floating mark is above or below the graph, an arrow on the edge points to it with its elevation, and the scale stays on the ground.
• 1:1 / V 1:n — the vertical scale: V 1:1.7 means heights are compressed 1.7 times to fit.

The bottom panel

The views that read the data in section share one panel under the view — under the photograph and the mono view in the mobile views. The view gets shorter to make room for it: nothing floats over the cloud, and the side panels keep their full height. Drag the edge above the panel to change its height; the height is remembered.

On its right, a column of buttons with an icon and a name chooses what it shows, one view at a time:

Every button keeps its place. One that has nothing to show here is grey, and its tooltip says what it needs. A view that appears — you open Multiview, compute a volume — comes to the front by itself.

The button at the bottom left of the mono view opens and closes the whole panel. It is orange while the panel is open, yellow while it is closed with something to show, grey when there is nothing. Closed by hand, the panel stays closed, even when a new view appears: that only lights the button. The same command is View → Bottom panel, and a keyboard shortcut can be assigned to it (Bottom panel).

The switches each strip used to have — View → Vertical profile, the Hide profile button in Multiview, Terrain → Volume profiles — now say whether that view comes forward by itself. Turned off, it waits in the column; pressing its button there turns it back on.

Changed. The vertical profile and the volume strips used to sit on the same edge of the window, over the view and over each other, and the station strip of the mobile views had no way to close. They are now the views of this one panel.

Options

• Radius — the area each view frames, 1 to 30 m, default 5 m. The camera never frames less than a floor distance, so a badly mis-placed point still shows the real surface around it.
• Pick the two viewpoints automatically — the bearing choice described above. On by default, persistent. Switching it off returns to fixed bearings (yours and yours plus 90°) and re-takes immediately — the way out if the choice ever misbehaves, without waiting for a new build.

Opening and closing

Multiview is docked, not floating: it takes the band beside the mono view, the same one the Zenith Overview uses, and it cannot be dragged or resized except by the splitter on its inner edge — which remembers how wide you left it. Because that band holds one panel at a time, opening Multiview closes the Overview and opening the Overview closes Multiview. That is the same rule the station strip and the sheet index already follow, and it is why the two share one button column.

Having no title bar, it has no close cross. It opens and closes from its button in the top-left corner of the mono view — see 8.6, The three window buttons — from View, or from its shortcut; where you left it is remembered, and it stays closed at the next start until you ask for it again. While the panel is closed that button turns amber and explains itself if vertical exaggeration is active, because opening Multiview holds exaggeration at 1.0× for as long as it stays open.

Gone in 4.0.0. Fix vertex has been removed: it did not work, and its job is done better by the pitch handles, which are always on the drawing and need no command. Confirm all buildings in draft has been removed for the reason given above. Elevation hysteresis has been removed together with the second, independent elevation sample it existed to rescue: a click now takes the elevation you can see (13.7).

9.3 The 3D cursor & elevation (Z)

While digitizing, the cursor carries a 3D object point (the point reported by the status bar, the stereo coordinatograph and the stereo reticle — always the same XYZ). You set that elevation by snapping to the cloud or by adjusting it manually; how the elevation behaves while you move the mouse in planimetry depends on which of the two cursor models is active (see 9.3.1).

You set elevation by snapping to the cloud or by adjusting it directly:

• Z — one-shot snap of cursor Z to the cloud under the pointer (also used by the Cloud Z object snap at pick time and by the trackball top-right button). Does not lock elevation for following vertices.
• Shift+Z — Z Anchor: sample the cloud under the cursor and lock that elevation for following vertices until pressed again (see 9.4).
• Ctrl+Shift+Z — toggle AUTO-Z (idle cloud elevation snap): when enabled, the cursor elevation follows the cloud surface after the pointer stops moving (~150 ms).
• Ctrl+Shift+Alt+G — toggle GND-Z (ground-biased Z for AUTO-Z; requires AUTO-Z ON). See 10.2.
• Ctrl+Shift+Alt+H — toggle WIRE-Z (collimate on the surface nearest the eye; requires AUTO-Z)
• Ctrl+Shift+Alt+A — Draw overhead line (two clicks on the wire; arms WIRE-Z with it)
• Ctrl+Alt+B — Build ground DTM (CSF).
• Ctrl+Alt+D — toggle Overview DTM layer.
• Ctrl+Shift+Q — toggle Constant cursor elevation (fixed world Z while moving planimetrically) vs floating-mark (constant view depth). See 9.3.1.
• Shift+wheel — nudge cursor Z by the fine step (default 0.01 units).
• Ctrl+wheel — nudge cursor Z by the rapid step (default 0.10 units).

The current cursor Z is shown in the Inspector, with a Reset Z control.

CAD geometry is fixed. Changing cursor elevation (wheel, Z, trackball ball) moves only the digitizing cursor — lines, polylines and other entities stay at their real elevation on screen. The CAD engine receives the cursor Z only when you pick a point (click or vertex confirm).

The mark under scattered canopy (3.89)

Digitizing on the cloud surface. When AUTO-Z is on or you are placing vertices with Cloud Z, StereoCloud3D first looks for a real, geometrically supported surface along the cursor ray: starting from the point nearest the camera within a small radius of the ray, it fits a plane through that point's own neighbourhood and accepts the first plane that at least half of its neighbours actually lie on. The elevation is that plane's intersection with the ray, so the mark is always under the cursor — obliquely as well as looking straight down — and a few scattered fronds or a thin patch of canopy in front of a roof no longer pull the mark onto them, because a scatter of points does not support a plane the way a roof does. Where nothing under the cursor supports a plane (dense unbroken canopy, or nothing in the cloud at all), the mark falls back to the nearest cloud point, the same behaviour StereoCloud3D always had.

Idle elevation readout. When no CAD command is active and AUTO-Z is enabled, the status bar and coordinatograph can show the cloud elevation under the cursor, but only after the pointer stops moving for a short moment (~150 ms). While you move the mouse, the readout does not chase every pixel — this keeps navigation smooth. When AUTO-Z is off, manual Z (wheel, one-shot Z, Z Anchor) is preserved while hovering. During an active draw/modify command, elevation follows the normal CAD cursor rules instead. A ring around the mono cursor (and, in stereo, around each eye's mark) shows how much to trust the current elevation: its colour runs continuously from red (no supported plane was found — the mark fell back to the nearest point) through amber to green as the fraction of the local neighbourhood that actually supports the plane rises. It is not a green/not-green signal — a thin margin above the acceptance threshold reads amber, a clean, well-supported roof reads green. Once locked, the WCS readout and stereo coordinatograph keep the same sampled point until the pointer moves again.

On the cardinal wall view (MainMonoUcs), the on-screen cursor is a compact CAD cross with pick box (matching native StereoCAD), not a full-viewport hair cross.

9.3.1 Constant elevation vs floating-mark

In an oblique georef view (MainMonoUcs), once you have set a cursor elevation and AUTO-Z / Cloud Z snap are off, StereoCloud3D can keep that mark under the pointer in two different ways. The choice is stored in settings. Mode B is the default: fixing the quota is the operator's choice.

How to switch

• Draw → Snap → Constant cursor elevation (fixed Z) — checkbox; shown with a tick when mode A is active.
• Z-FIX on the Object Snap toolbar, docked or floating, next to EDGE — the same switch at one click, lit while mode A is active. Its tooltip gives the name, the shortcut and what it does.
• Ctrl+Shift+Q — toggle (rebindable in Keyboard Shortcuts).
• The status line reports which mode is active after each toggle. The choice is saved between sessions.

When each mode applies

9.4 Object snaps

Open Draw → Snap or the Object Snap toolbar to enable running snaps. A master Object Snap toggle turns them all on/off. Available modes:

EndpointMidpointCenter IntersectionNearestPoint PerpendicularTangentCloud Z

Cloud Z snaps a vertex's height to the point cloud at pick time — the key to fast, accurate restitution. Like the other running object snaps, it is applied when you click or confirm the next CAD point. When the cloud pick succeeds during drawing, the mono cursor shows a green ring and the sampled elevation as confirmation.

Locked Z (Shift+Z) takes the elevation of the cloud point under the mono cursor and gives that exact Z to every following vertex until Shift+Z is pressed again or the Z-LOCK button in the snap bar is clicked. While it is on, the cursor stays at that elevation as it moves, the wheel does not change it, AUTO-Z and Cloud Z object snap are suspended, and a red anchor marks the sampled point. The locked elevation is shown where you look: the Z-LOCK button is lit, the status bar reads Z LOCKED with the value, and the stereo coordinates and the cursor carry the same number. Use it for repeated points that must share one quota. (Release 3.19 and earlier used bare Z; upgrade users should reset keyboard shortcuts if an old Z binding still toggles it.)

9.4.1 What the snap aims at (Cloud / Vectors / My choice)

The modes above are independent switches, so declaring “right now I want the vectors” used to mean turning several of them off one at a time. Three entries at the top of Draw → Snap do it in one click:

• Cloud — snapping to the cloud surface only. The entity snaps go off.
• Vectors — snapping to drawing entities only, and it brings back your entity snaps, not a set chosen for you. Only if you have never picked any does it start from the CAD default.
• My choice — the combination you built by hand, returned exactly as it was.

These are presets over the modes below them, not a new mode: nothing is added to how snapping works. My choice is what makes the other two usable — switch to Cloud for a roof pitch, come back, and your configuration is there again instead of having to be rebuilt one switch at a time. Any mode you toggle by hand becomes the remembered combination, and so does Restore defaults.

They live in the menu and not on the toolbar on purpose: they set the very switches the bar already shows, and sitting directly above them they read as a second, competing control rather than as a shortcut over the first.

Arming Cloud against a hidden cloud is refused, as arming Cloud Z alone always was. If your own combination contains Cloud Z and the cloud is hidden, My choice comes back without it and says so, rather than restoring a snap that cannot bite.

9.4.2 Which coordinates the snap imposes (XYZ / XY / Z)

A snap normally hands the vertex all three coordinates of the point it found. In survey work that is often more than you want: you may need the planimetry of an existing vertex at the elevation you are already digitizing at, or the elevation of an eave line while you click the planimetry freely. The filter beside the OSNAP button decides which of the two an object snap is allowed to write:

• XYZ — the snapped point as the CAD engine returns it. Planimetry and elevation both come from the entity. This is the default and the usual CAD behaviour.
• XY only — the snap gives the planimetry; the elevation stays the one that governs the cursor (cloud, Z Anchor, wheel). Trace over an existing drawing while keeping your own quota.
• Z only — the snap gives the elevation of the entity under the pointer; the planimetry stays where you click. Draw at the height of an existing eave, ridge or cable without being pulled onto it. Available in the georeferenced 3D view, the only one where elevation and planimetry come from different sources.

The button shows the state in force — XYZ, XY or Z — and each click moves to the next one. It is lit whenever a filter is actually in the way, and amber when Z only is chosen in a view that cannot apply it. The same three states are in Draw → Snap → Coordinates the snap imposes, on Ctrl+Alt+X, and on the command line as SNAPXYZ, SNAPXY and SNAPZ. Your choice is saved between sessions.

With Z only, hovering an entity moves the cursor readout to that entity's elevation while the mark stays under the pointer: what you see is what the click will write. Move away from the entity and the elevation returns to whatever was driving it before — AUTO-Z, the anchor or the wheel.

9.5 Polar tracking

Polar tracking helps while drawing lines and polylines: when the next segment is near 0° or 90° relative to the previous segment, a dashed guide appears and the cursor magnetizes to that direction. This is separate from object snaps (perpendicular to an existing entity is a snap mode, not polar tracking).

• POLAR checkbox on the Object Snap toolbar, or Draw → Snap → Polar tracking.
• F8 — toggle polar tracking on/off (also PON on the command line).

Default: on, relative angles in 90° steps. Your choice is saved in settings between sessions, and it applies to whichever drawing you are working on — the georeferenced 3D drawing as well as a sheet. The angles are planimetric (world XY): polar fixes the direction of the next segment, while its elevation stays whatever Cloud Z or AUTO-Z is giving it. A small on-cursor tooltip reads out the polar distance and angle.

9.6 Pointing devices

Every device you point and navigate with has its own settings window under Tools → Pointing devices: Mouse, Trackball settings and SpaceMouse. Each is named after the window it opens.

Trackball (Kensington / Logitech)

StereoCloud3D supports professional trackballs for 3D cursor Z, view rotation, and quick actions. Recommended: Kensington Expert Mouse (wired) or Expert Trackball (wireless) — large ball, scroll ring, four buttons (USB 047D). Also supported: Logitech ERGO M575 (successor to the M570, still sold new), Wireless Trackball M570 and Trackman Marble (discontinued but widely available refurbished/used; USB 046D). Use any of these with a normal mouse for pan and CAD pick.

• Ball — raises or lowers the 3D cursor Z (the elevation quota used while digitizing). The trackball cannot move the pointer while the ball turns, so the view does not jitter; your other mouse still moves it, so you can change Z and XY at the same time.
• Scroll ring (Kensington) or wheel (Logitech) — changes the Z action speed (shown in the trackball settings; no longer drawn on the stereo coordinatograph).
• Mouse MMB + ball — hold the middle button on your mouse (without dragging the mouse) and roll the trackball with the other hand. Rotates cloud and CAD live around the current orbit pivot (yaw/pitch), using the same native CAD/cloud orbit host as Shift+MMB.
• Alt + ball — same rotation if you prefer a keyboard modifier.
• Hardware buttons — Kensington: four assignable buttons (defaults below). Logitech thumb models (M570/M575): left/right map to the bottom-left / bottom-right slots. Trackman Marble: four buttons (same assignment grid).

Open Tools → Pointing devices → Trackball settings to enable the device (trackball support is off at startup until you turn it on here — improves stability on some PCs), invert ball/ring sense, reset the speed multiplier, and assign buttons. Default assignments:

• Top-left — pan the view to the cursor XY (one click; recenters while keeping the world point under the cursor).
• Top-right — snap Z to the cloud surface under the cursor.
• Bottom-left — toggle cloud visibility.
• Bottom-right — toggle CAD visibility.

Preferences are saved in settings.json (kensingtonTrackballEnabled, speed, invert, and the four button assignments).

Mouse

Tools → Pointing devices → Mouse holds the settings of the ordinary mouse:

• Wheel zoom speed — how much one wheel step zooms (0.5 to 5). The SpaceMouse zoom builds on it.
• Cursor height fine step / rapid step — how far Shift+wheel and Ctrl+wheel raise or lower the 3D cursor per step, in project units.
• Invert wheel Z direction — for operators used to the opposite direction on another plotter.

SpaceMouse (3Dconnexion)

A 3Dconnexion SpaceMouse lets you work with two hands: the left hand on the SpaceMouse moves the view, the right hand keeps the mouse on the stereo cursor to draw. Cloud, CAD and both stereo eyes move together, exactly as with the mouse, and the cloud stays at its navigation detail while the view moves.

StereoCloud3D uses the SpaceMouse in one of two ways, chosen by itself when it starts; the log says which ([spacemouse] path=navlib or path=hid).

With the 3Dconnexion driver (3DxWare 10)

When the 3Dconnexion driver 3DxWare 10 (version 10.8.12 or newer) is installed, StereoCloud3D works through it. There is nothing to calibrate or set in StereoCloud3D: speeds, axes, dominant axis and buttons are set in 3Dconnexion Settings, the same way as for every other program, and StereoCloud3D appears there under its own name. The SpaceMouse window (Tools → Pointing devices → SpaceMouse) then shows only whether the SpaceMouse is switched on, and says that the driver is in charge.

• Push the cap sideways or forward/back — pan.
• Press down / pull up — zoom around the mouse cursor, like the wheel.
• Twist and tilt — rotate the view, the same orbit as the right-button drag. As with the mouse, the orbit needs the cloud under the mouse cursor.
• The device keys (Fit, views, menus) are run by the driver.

3D input device development tools and related technology are provided under license from 3Dconnexion. © 3Dconnexion 1992 - 2025. All rights reserved.

Without the driver: direct reading

Without 3DxWare StereoCloud3D reads the SpaceMouse directly. Plug in its receiver (or cable) and switch it on: StereoCloud3D finds it within two seconds. The SpaceMouse window says that the driver was not found and that the settings below apply.

With the default settings:

• Push the cap sideways or forward/back — pan.
• Press down / pull up — zoom around the mouse cursor.
• Tilt forward/back and twist — rotate around the point under the mouse cursor, the same orbit as the right-button drag.
• Keys: Menu opens the SpaceMouse window, Fit frames the scene, T returns to the plan view, the rotation lock key switches rotation on and off (the status line says which).

Tools → Pointing devices → SpaceMouse makes everything adjustable:

• Cap axes — each of the six movements is assigned to a function (pan sideways, pan up/down, zoom, rotate, tilt the view, or nothing), with its own Invert and Speed. The Now column shows each axis live, so you can see which movement is which.
• Dead zone, Response (1 = linear, higher = finer with a light touch) and Only the strongest movement, so that a pan does not also rotate by accident.
• Keys — every key can run any program command, chosen from a list you can search by typing. Press a key on the device and its row lights up; while the window is open the keys only light up, except Menu.
• Restore defaults brings back the assignments above.

9.7 Input recovery & view realignment

Release 3.20 adds recovery commands for sessions where the mono cursor disappears, navigation keys (T, Z) stop responding, or pan/orbit feels stuck (often after Alt+Tab or focus loss during MMB drag).

• Unlock input (Ctrl+Shift+R) — View → Unlock input or the Recover input shortcut. Ends active pan and orbit drags, cancels UCS-orient preview and orbit-pivot sessions, closes escapable modals (wizard, new-project dialog, etc.), invalidates cursor caches and reclaims keyboard focus on the main window. Does not change your current orientation, zoom or framing.
• Restore view and alignment — View menu only (no default shortcut; assign in Keyboard Shortcuts if needed). Runs Sblocca input then the same full zenith reframe as T (Top / WCS): resets oblique orient to R=I, re-syncs cloud, CAD and stereo camera. Use when cloud and CAD have drifted apart after a partial orbit or pan — equivalent to a controlled T after unlocking input.

If you lose window focus while panning (MMB) or orbiting (G drag), StereoCloud3D automatically ends the stuck drag when focus returns — without resetting the view.

9.8 Measuring on the cloud

A consultation tool: it reads coordinates off the point cloud and shows numbers. It writes nothing — no CAD entity, no project data, no saved object. Closing the tool leaves the project exactly as it was.

Neither command has a keyboard shortcut by default; assign one in Keyboard Shortcuts if you use them often. The tool cannot be started while a CAD command is running, and it never takes a click away from a CAD or cloud tool that is already active. If it refuses to start — no cloud, or a CAD command still running — it says so in the Command Line.

Where the numbers appear

Starting either command opens the Measurement results panel, and that is where the measurement is read:

• Point info — X, Y and Z of the picked point, in project coordinates, to three decimals.
• Distance — both points in full, then 3D distance, Horizontal and Height difference. The height difference is signed: it is the second point minus the first, so it tells you which way the ground goes.
• Copy — puts the numbers on the clipboard, tab separated, so they paste into a spreadsheet as columns and into a report as text.
• Measure again — starts a fresh measurement of the same kind without going back to the menu.
• Clear — ends the tool, exactly like the menu item and like right-click.

While the tool is waiting for a click the panel says which click it is waiting for. Closing the panel does not end the measurement: Cloud → Measure → Measurement results brings it back with the numbers still there. Each result is also written to the Command Line, so a session of several measurements leaves a readable list.

The two points and the line between them are drawn on the main (mono) viewport, with the same numbers repeated beside the last point so you can read them without leaving the cloud. They do not appear in the external stereo output.

If there is no cloud fragment under the pixel you clicked — a gap between points, or background — the tool says so and waits for another click.

9.9 Navigation trajectory

Draw a control polyline on the cloud and the camera will fly along it, at a speed you set in metres per second. The view axis stays horizontal, with heading along the path tangent (plan projection): grades change flying height, not pitch. Navigation itself runs in stereo. The trajectory (cyan/yellow) is drawn on the mono viewport while you author or pause it; during Play it is hidden so it does not sit in the stereo pair.

Open View → Toolbars… and tick Navigation.

1. Draw camera path... — if there is no ground DTM yet, StereoCloud3D builds one first (CSF on the Cloud AOI, or on the current view). Then click the path on the ground; each vertex is draped on the DTM. Right-click to finish. Two points are enough, three if the path is closed. Undo point, or Backspace, removes the point you just placed.
2. Closed — the path joins its end back to its start and the travel is continuous. Left open, reaching the end restarts from the beginning.
3. Speed — metres per second. The camera covers that distance along the path itself, so it does not speed up where the control points are far apart.
4. Offset from ground — metres above the DTM. The drawn path stays on the terrain; this offset is the flying height of the camera.
5. Look left/right — yaw while the flight is running (or paused). Negative looks left of the path, positive looks right; 0 follows the tangent. The view stays horizontal. Use the slider or the Left / Right arrow keys (5° per press; hold to keep turning). Stop returns the offset to 0. This look is not saved with the path.
6. Play, Pause, Stop, Restart, and the direction button to travel the other way. The slider shows and sets the distance travelled.

Pause and Stop are different on purpose. Pause holds the camera where it is, so you can look around and use Point info and Distance from that viewpoint, then carry on. Stop ends the flight and hands the camera back to the normal view exactly as you left it.

The drawn path is saved with the project, in navigation/trajectory.json under the project folder, in project coordinates. Dialog position, open/closed state, speed and ground offset are remembered in the application settings so they survive closing the project and restarting StereoCloud3D. Opening a project that already has a saved path still uses that file's speed and offset for that project.

Flying the trajectory changes nothing but the viewpoint. The drawing, the technical stereo view and all CAD behaviour are untouched, and CAD restitution stays refused while the Navigation view is on.

10.0 Cloud classification

Use Automation → Classification to write ASPRS classes into the loaded cloud octree (in place). The Automation menu sits between Terrain and Stereo; it also holds the building-outline commands (13.11). The submenu starts with Classify cloud… (the whole chain, below) and then lists the single passes, ordered by the recommended run sequence:

1. Ground (class 2) — SMRF morphological filter; required before most other passes.
2. Buildings (class 6) — roofs measured against class 2.
3. Roads (11) / Water (9) — surfaces among ground points.
4. Masts and poles (15) / Overhead wires (14) — vertical features on unclassified points (mobile surveys; an aerial survey asks for its power lines by voltage, see 10.0.0c).
5. Vegetation (3/4/5) — run last on remaining unclassified points.
6. Mark noise (7/18) — flags outliers; nothing is deleted. The two noise classes are hidden by default: every time a project opens the class list shows all classes except 7 (low point / noise) and 18 (high noise). Press All on in the Display panel class list to show them; your choice is kept until you change project.

With several cloud layers, each pass opens a submenu to pick the layer. After a pass, reload colour mode Classification in Cloud Visualization to see the result (RGB colouring hides classes). Commands are also listed under Keyboard Shortcuts (no default key bindings; assign if needed).

10.0.0a Saved classifications — choosing one, and throwing it away

Every classification run saves the result as a version, listed in the Display panel under Classification version and in the Scene tree under the cloud's Classification folder. Choosing one rewrites the class byte of every point, so the cloud goes back to what that run decided — it is the data that changes, not just the colouring.

With several cloud layers the list has one group per cloud, under the cloud's name, and each group marks its own active version: a version always belongs to the cloud it was taken from, and choosing it rewrites that cloud only.

A version costs one byte per point: about 17 MB on a 226-million-point cloud, and a run leaves one behind every time. To remove one, press Delete next to it and confirm. It cannot be undone: those bytes are the only copy of that run's answer.

Two are never offered for deletion:

• the active one — deleting it would leave the label pointing at bytes that no longer exist. Activate another version first, then delete it;
• Classification — Original, the backup taken before the first pass ever ran. It is the only record of how the cloud arrived, and a delivered classification cannot be recomputed.

10.0.0b Change class — correcting the points you choose

The automatic passes get most of a cloud right; Automation → Classification → Change class… is for the rest. It opens a window where you say which points and which class:

The choices are remembered from one session to the next. Then say where, in one of two ways:

• Draw polygon on view — click the vertices on the view with the left button; right click or Enter closes the polygon and changes the class. Backspace or Esc removes the last vertex; Esc on an empty polygon stops drawing. Every point that falls inside the polygon as you see it on screen changes, hidden points behind the visible surface included — in a plan view that is everything above and below, in an oblique view everything along the line of sight. Use From classes to keep what you do not want to touch. The polygon stays attached to the cloud while you pan and zoom with the middle button; after closing one you can draw the next straight away.
• Use selected CAD polylines — select one or more closed polylines in the drawing first. Every point inside them in plan changes, at any elevation. Open polylines and other entities in the selection are skipped, and the window says how many when none of the selection can be used.

The change runs like a classification pass, with the same progress window, and the cloud redraws by itself when it is done. The status line says how many points changed and how many were inside the area; the Command Line lists them by the class they came from.

While a polygon is being drawn the right button closes it, so right-drag does not orbit; navigate with the middle button. The command has no default key: assign one in Keyboard Shortcuts (Change class).

10.0.0 Classify cloud — the whole chain in one run

Automation → Classification → Classify cloud… runs every pass in the required order, so you do not have to remember it. The dialog asks for four things:

The run is a background job with a progress list, one line per pass and its point count; the Command Line keeps the same record. Nothing is deleted at any point: noise is flagged, and you hide it from Cloud Visualization.

10.0.0c Overhead lines — high and medium voltage (aerial surveys)

On an aerial survey the wires of a power line are a handful of returns per metre, often with gaps, and no drawing to follow. The program finds them from the points alone, with no seed clicks and no classes, and tells the high-voltage line from the medium-voltage one: each gets its own layers in the drawing and its own point classes. There are two ways in, and they share the same search:

• Classify cloud — the boxes High-voltage lines (14, 13, 15, 16) and Medium-voltage lines (23, 24, 25). Both start ticked on an aerial survey; with both unticked nothing is searched and no class is touched.
• Reconstruction → Power lines → Automatic search (tab Power lines, or the Search button of a delivery row): finds the lines and shows them as a draft on the cloud, to check before anything is written to the drawing.

What the search does

• Reads the cloud in tiles (256 m), so a whole survey is never loaded in memory. It works on airborne surveys as well as on mobile runs; on an aerial survey the progress bar shows the tiles read and the phase (Finding the clean wires, Finding the bundles of wires close together, Joining the spans across the supports, Following the wires through the holes).
• Keeps only the points that can be wire: well above the ground, isolated in the air, lying along a line.
• Fits catenaries: a wire is straight in plan (piece by piece), its heights follow a catenary, and where returns are missing it bridges the gap with the same curve, so the two sides join precisely. A second, more permissive pass finds the bundles of wires a metre or so apart; those lines are marked to review.
• Continues the wires through the holes: at the middle of each piece found, a section across the wires is drawn automatically and the search by section (15.12) follows them in both directions, also across gaps with the model of the span. A point counts only if an isolated candidate lies near it. The continuation stops at angle poles; the next piece found goes on from there.
• Chains the spans across the supports and looks for the column of points that justifies each support. A support exists only where the spans meet and a column of points justifies it: the column is justified when at least 60% of its 1 m slices, from the ground to the attachment, hold a point. The ends of lines are candidates too, but a support must be at least 8 m tall, because hedges and trees at the end of a line have perfect columns at 3–7 m. A pole (under 15 m) is a thin column with its foot in the grass and is measured as such. A support with no justifying column is not drawn: it is counted as junction without a column of points.

Sensitivity

The combo Sensitivity in the tab Power lines sets how eager the search is: Cautious (only the clean wires; steps over interruptions up to 8 m), Normal (up to 12 m: a tree across the line does not stop it) and Sensitive (up to 15 m: finds wires deeper in the trees, with a few long false lines to remove). On an aerial survey Normal and Sensitive also look for the bundles of close wires. The choice is remembered. In the same tab Remove deletes a wire or a support of the draft with a click and Undo remove puts it back; Confirm and Discard are as above.

High voltage or medium voltage

Each line is decided by two measures: the median length of its spans and the height of its wires above the ground. On the survey the thresholds were measured on, high-voltage spans were 342–345 m with wires 16–23 m up, medium-voltage spans 23–112 m with wires 7–11 m up. The two measures must agree; when they do not, the line is not guessed: it goes on its own layer, Lines to review, and the summary counts it. Something whose wires hang less than 4 m above the ground is not a power line and is not drawn (the Command Line says why). A support takes the voltage of the line it stands under.

The layer names are in the interface language (Linee AT - conduttori in Italian, and so on). The classes 23–25 are new and belong to the standard legend, in blue against the orange of the high-voltage family; classes 19–22 stay free for what your customer decided they mean.

The classes 13–16 are all orange, in four tones so the parts can be told apart on the cloud: guard wire (13) light, conductor (14) the base orange, insulator (16) darker, tower (15) darkest. When a line is classified, the program also accepts uncertain lines whose spans are measured for more than 70% of their length, which recovers more of the medium-voltage wires.

Checking the draft, then confirming

The result is a draft drawn on the cloud: orange high voltage, blue medium voltage, amber to review. Remove deletes the selected line (Undo remove gives it back), Confirm writes the lines and supports to the layers above, Discard throws the draft away. Nothing reaches the drawing before you confirm, and pressing Search again repeats the search from the start.

Classifying with the lines

When Classify cloud runs, each ticked voltage uses what is already drawn on its layers (found earlier and kept, or corrected by you) and searches only where there is nothing. The high-voltage pass goes first; the medium-voltage pass leaves alone every point the first one set, so a shared support or a crossing is never taken back. Wires that are scattered get a wider tube (60 cm), but only for points that are isolated in the air, so nearby vegetation is not taken. At the end the summary says how many points each pass really changed — a point given the class it already had is not counted — and the Command Line lists them class by class. The noise classes (7 and 18) are hidden by default; show them from the class list when you need them.

10.0.0d When the lines are green, and reviewing the ones to check

In the Deliverables view the power-line step is green only when it has been measured, never because the search ended. For every wire of the draft three criteria are checked:

• Residual: the 95th percentile of the vertical distances between the points and the fitted wire must be within the tolerance of the type of work for conductors (15 cm);
• Deduced share: no more than 5% of the length of a wire may be a stretch crossed with no points (the model bridged it, nothing was measured);
• Columns at the joints: at every junction between two spans there must be a column of points that justifies the support.

A wire that was not measured (drawn by hand, or a continuation) and a line the voltage measures did not decide also go to review. If any criterion fails, the step is yellow and the summary says which one and by how much: To review: n doubtful line(s); n out of tolerance (worst 95th percentile n cm against n cm, median n cm); n with more than 5% of their length crossed with no points (n m); n junction(s) without a column of points; n not measured. The 15 cm and 5% limits were measured on one survey (5 high-voltage and 22 medium-voltage lines) and may need to be set again on others.

After the draft is confirmed, the step stays yellow as long as the layer Lines to review holds anything, and is green only when it is empty. Under the yellow step the button Review the n to check starts the review: the view goes to the first element and selects it, and a small window Lines to review asks Is it part of a high-voltage or a medium-voltage line? for a wire of n m or a support, n m high:

• High voltage / Medium voltage — the element is recreated on the layer and with the colour of that voltage;
• Delete — it is not a power line;
• Next — leave it for later and go to the next.

At the end: Review finished: n moved to a line, n deleted, n left to review.

10.0.0e Span and support tables, ground profiles

Once the lines are confirmed, two steps of the chain write the delivery tables. Each runs with a progress bar and writes a .csv with an .xlsx copy beside it, in the exports folder of the project (the column titles are in the interface language, the decimals use the dot):

The ground profiles need a ground model: No ground model: classify the cloud first (the Ground step). Without it the step stops and says so: it does not invent a terrain. A station outside the ground model is written -. After restarting the program the Deliverables view counts the rows of the files again.

Chainage and offset from an axis. When the type of work takes the axis of the line from a drawn line, the support inventory table of the project (sostegni.csv) has the columns axis chainage m and offset from axis m (+ right). The axis is the only open polyline on the layer ASSE; the offset is positive to the right of the axis, looking along it (it was positive to the left before this release). If the layer has no polyline, more than one, a closed one, or one with fewer than two distinct vertices, the program says so in the Command Line and in the status bar and writes the two columns as - (never 0).

10.0.0f Earth wires and the classes

The earth wires (guard wires) of a high-voltage line are the points of class 13: in the Deliverables view the row Earth wires is done when class 13 has points, there is no layer to draw. The four classes of a high-voltage line now share the same orange hue in four tones, so they can be told apart on screen: earth wire (13) light, conductor (14) orange, insulator (16) darker, tower (15) darkest. On an aerial survey with no wires drawn, Classify cloud searches the wires by itself with the same engine (it no longer uses the generic wire and support passes, which are tuned for mobile data), and places the supports from the wires; it also admits the uncertain lines that are measured for more than 70% of their length. The medium-voltage classes (23, 24, 25) are blue as before.

10.0.0g Class comparison — how far two classifications agree

The bottom panel has a Class compare view that compares two saved classification versions of the same cloud, point by point. Use it to measure a new run against the classification delivered with the data, or against another run. It appears (as a tab of the bottom panel) when a cloud layer has at least two versions; until then the tab is greyed with that reason. Nothing in the cloud changes.

1. With several clouds, pick the Cloud layer.
2. Choose A and B. A is the reference; B is the version measured against it. By default A is the original delivery and B is Current (in the cloud). The versions are listed with (active) next to the one in the cloud. Swap exchanges them. The choice is remembered for each project and layer. Choosing the same version twice is refused.
3. Press Compare. A wait window shows the progress and Esc stops it (no partial result is kept). A version that belongs to a different cloud index is refused whole, never compared.

The result is a summary line (agreement %, points compared, points that differ, points not compared) and two tables:

• Per class, with the names and colours of the current legend: In A and In B (points of the class in each version), Precision (of the points B puts in the class, the share A also has there), Recall (of the points A has in the class, the share B keeps) and F1 (the two in one number, 100% only when B equals A on that class).
• Where the points of A go in B: a matrix with A in the rows and B in the columns. The diagonal is agreement. Hover a cell for the two class names, the number of points and the share of the class in A. With too many classes the matrix is replaced by a message.

Show differences on the cloud draws the points whose class differs in magenta on the mono view (a uniform sample of up to 25,000 points per pair of classes). Click a cell off the diagonal to show only that pair (for example A ground, B vegetation); click it again, or press All differences, to show every pair again. If the cloud's classification changes after the comparison, the panel tells you that it is out of date and to compare again.

10.0.0h Export classified cloud (LAS / LAZ)

Cloud → Export classified cloud… writes the classes the cloud holds now into a LAS/LAZ file you can deliver. What is written is what the cloud file holds on disk, that is the active classification version; changes not stored in the cloud are not included. The dialog shows which classes are being exported.

• Normal export (from the original). The program rereads the original LAS/LAZ the cloud was imported from and changes only the class of each point; everything else in the file is copied as it was. The project does not remember where the originals are, so you choose them with Choose the original file(s)… (the dialog shows where the cloud was imported from). Before anything is written the files are checked against the cloud: each must be one of the files it was imported from, it must be readable as LAS/LAZ, and the total number of points must equal the number at import. Points left out of the cloud at import keep the class they had in the original.
• Output folder. Each output has the same name as its original. An original is never overwritten: the folder must be a different one, and the dialog refuses a folder that holds one of the originals.
• Declared alternative (no original). Only when the original is not available, tick Declared alternative: no original file, export from the cloud only. The file is then rebuilt from the project cloud, which keeps only position (to the millimetre), class, and intensity and colour reduced to 8 bits. The dialog lists what is missing from the result: return number and number of returns, GPS time, scan angle, point source id, the full 16-bit intensity, extra fields, the coordinate system and the original header; points left out at import are absent too. It is never chosen for you.
• A wait window shows the phase (reading the classes, reading the original, matching the points, writing). At the end the dialog gives the points written, and in the normal mode how many changed class and how many were unchanged.

In the delivery chain the Classified cloud step runs this export by itself, always in the normal mode: it looks for the original among those you named before and in the folder the survey data came from, and only accepts one with the same name and the same number of points. The result goes to exports/nuvola_classificata in the project folder, and after any successful export the project records where the file went, so the step shows as done wherever you saved it. When no original is found the step stays red with the reason and its button opens this dialog; the declared alternative never starts on its own. The command has no default key (Export classified cloud in Keyboard Shortcuts).

10.0.1 Surface models (DTM / DSM / nDSM)

New in 5.2: Build surface model (DSM). In the menu of the Terrain activity. The highest point of every cell, in the area of interest if there is one, otherwise on the whole cloud; 0.5 m cell, wider only on very large clouds. It writes <layer>/surfaces/dsm.tif (the same file as the classification: running it again replaces it) and shows it in the overview. Progress bar, and Cancel (Esc) stops it.

A whole-chain run (Classify cloud) does not only write the class byte. Once the passes are done it writes three georeferenced Float32 GeoTIFFs into <layer>/surfaces, at 0.5 m cells:

Empty cells carry NoData, never 0, which would read as sea level. No coordinate system is attached: the rasters carry the correct world geotransform and nothing more, so tell your GIS which system the survey is in.

Each raster is filed in the Scene Tree under Products → Surface models, with the run that made it, its cell size and its path. Re-running the classification overwrites the same files and updates the same three entries rather than stacking duplicates.

Looking at one. Tick a raster's box in the tree, or right-click it and choose Show in the Zenith Overview. It is read off disk in the background and shaded under the CAD in the Zenith Overview, whose header names the raster on display. There is one raster layer, so showing a second one replaces the first. A block wider than 2500 cells is averaged down for display; the log line says by how much.

The DTM is also a measuring surface. Showing dtm.tif makes it the session terrain, so GND-Z samples it — the same state as a Build ground DTM (CSF) run, without the wait. A classification run adopts its own terrain automatically for the same reason. A DSM or nDSM is only ever displayed: it is not a ground reference, and GND-Z keeps snapping to the terrain.

10. Cloud visualization

Open Cloud → Cloud Visualization… (Ctrl+Shift+L) for point color mode, point size and related display options. Settings are saved with the project (and in app settings where applicable).

Color modes include RGB, elevation, intensity and Classification. In Classification mode an ASPRS legend lists standard LAS classes; use the per-class checkboxes to show or hide points (for example ground only, or buildings without vegetation). Classes with no points in the loaded tiles remain listed but contribute nothing until classified data is present.

10.1 Color modes

Modes whose attribute is missing in the data are shown as unavailable.

10.2 Elevation range (automatic, spike-resistant)

Elevation coloring needs a min/max height. Real clouds often contain a few spikes (a stray point far above or below the surface) that would otherwise wash out the ramp. StereoCloud3D therefore computes a robust range from the 2nd–98th percentile of the height distribution, so the colors span the meaningful terrain rather than the outliers.

10.3 Point size & cache

• Point size — on-screen size of each point (1–64).
• Cache tiles — how many tiles stay resident in memory (default 32). Larger values reduce re-loading at the cost of memory.

Fast cloud rendering

Point clouds indexed as an octree are drawn from buffers that stay on the graphics card: each node of the tree is uploaded once and kept there, so changing the view only re-issues the draws instead of rebuilding and re-uploading every visible point. Navigation stops costing in proportion to how many points are on screen, which is what makes rotation and zoom on a large cloud feel immediate.

There is nothing to switch on: it is how an octree cloud is drawn. A project built on grid tiles keeps the classic path. CAD snapping, cloud picking, measurements and the tile data are the same either way.

If the cloud misbehaves on a particular machine — a graphics driver, an unusual card — support can start the application once with the environment variable SC3D_CLOUD_RESIDENT_GL=0, which restores the classic rendering path so work can continue while the cause is found.

How much detail actually reaches the screen (4.7)

For each view the renderer chooses which nodes of the octree it needs, and then holds those nodes ready to draw. Until 4.7 those two quantities were governed by different limits: the choice could run up to 40 000 nodes while only 16 384 could be held, and the difference was quietly set aside. In an oblique view at middle zoom — where the line of sight runs along the ground and collects nodes at every distance — that meant 59% of what had been chosen never appeared, and the depth of detail stopped two levels short of what the data contains.

The two limits are now the same number, so a node that was chosen is a node that is drawn. There is nothing to set. A large cloud in an oblique view uses roughly a third of a gigabyte more memory and draws at the same rate; while the extra nodes are being read for the first time you may notice a brief pause, once.

The same mismatch was what stopped the stereo image in oblique views. The compositor waited for the set-aside nodes to arrive, and they were never going to: they had been given up on, not queued. In a zenith view, or zoomed close in, the selection stayed under the limit and the freeze never happened — which is why it looked like a stereo fault rather than a cloud one.

10.4 Gaussian splatting and Eye-Dome Lighting

Two visual enhancers for dense point-cloud interpretation, both reachable from the Representation dialog (and still from the Cloud → Performance menu):

• Gaussian splatting draws each point as a soft, screen-space Gaussian splat that blends with its neighbours, filling the gaps between measured points into a continuous surface. It improves apparent surface continuity without changing CAD snapping, cloud picking or the underlying tile data. (It is the evolution of the earlier “Pro surfel visual overlay”.)
• Eye-Dome Lighting applies a depth-aware post-process to the cloud texture. It darkens local depth discontinuities so edges, steps, cavities and façade details read more clearly even when the cloud has no normals.

EDL is visual-only. It is applied per cloud framebuffer (mono and each stereo eye) before CAD vectors are composited, so CAD lines remain crisp and are not shaded. The controls are:

10.5 Representation dialog

Cloud → Representation — also reached by right-clicking the cloud in the 3D view — gathers the point cloud's display controls in one place: the color mode, point size and the two enhancers above (Gaussian splatting and Eye-Dome Lighting). It is the cloud-only part of the full Display dialog, brought within one click of the cloud you are looking at, so you can change how the points read without leaving the view.

The classification, intensity and RGB colouring described above are untouched by these two toggles: Gaussian splatting and EDL change how the points are drawn, not what colour carries the data.

10.1 Zenith Overview

View → Zenith Overview (Ctrl+Shift+O) docks a WCS plan panel on the left of the mono viewport, separated by a draggable vertical splitter (the split ratio is remembered between sessions). It always shows CAD in world coordinates (plan / zenith), preferring georeferenced imagery under the drawing when available. The main mono viewport keeps focus for digitizing and can remain oblique while the overview stays top-down.

• Fixed crosshair, moving map — a crosshair stays fixed at the centre of the panel while the imagery and CAD map follow the live mono cursor. The scale matches the mono view exactly: a metre is the same number of pixels in both panes, so an entity is the same size in the overview as it is in mono. The overview therefore covers less ground than mono whenever it is the smaller panel — it is a window at mono's scale, not a reduction of the whole view. This is what you get while Follow mono (below) is ticked.
• Follow mono — a tick box in the overview header, on by default, which is the behaviour described above. Switch it off and the overview keeps its own centre and scale: it no longer moves when the mono view moves, so you can study one detail of the DTM or of the imagery without disturbing the mono view. Turn the wheel to zoom about the pointer and drag with the left or the middle button to move the map. When you switch it off, it starts from the framing you have on screen. The choice is remembered between sessions. Imagery keeps streaming for whatever the overview is showing.
• Live entities — entities you draw in the mono CAD appear in the overview: its plan drawing is refreshed from a snapshot of the live drawing whenever the entity count changes, without disturbing the mono view.
• Imagery — while the overview is open, WCS imagery continues streaming even if the mono view is oblique, and tiles fill the overview extent.
• Commands in the header (5.8) — what is under the overview is chosen in the overview's own header row, always there while the overview is open: the raster drop-down (None, or one of the project's surface rasters — one at a time; the one used as ground terrain says so), its opacity, how the terrain reads (Shaded relief, Elevation, Slope, Aspect; hover the drop-down for the colour legend), In mono (the same raster under the cloud in the mono view, see 10.1.1), Imagery when the project has some, and Follow mono. When the overview is narrow the row wraps instead of cutting a command. These commands are no longer in the Display panel. The ground DTM is built with Cloud → Build ground DTM (CSF)….

Use the overview as a situational map while digitizing walls or roofs in an oriented UCS. Drag the splitter to resize the panel.

10.1.1 Orthophoto and DTM in the mono view

The mono view can show the same orthophoto or terrain model (DTM) that the overview shows, in place of the bare points. This is offered only when the mono view is looking straight down: in that view a flat raster and the cloud cover exactly the same ground. The view may be rotated about the vertical — a plan turned a few degrees is just as exact as a straight one. In an oblique or perspective view, and in stereo, a flat picture would not match the cloud, so nothing is shown and the controls disappear.

The quick bar. When the view looks straight down and the project has something to show (a terrain model, a surface raster or imagery), a small bar appears at the top centre of the mono view. Its buttons:

The raster is drawn opaque: there is no opacity slider, because a see-through raster adds nothing to what you read. The CAD drawing is always on top of both raster and cloud. The DTM is drawn in the way chosen in the Display panel under How the terrain reads.

The same switch exists as a tick box in the Display panel (View → Display, Overview tab, Surface section): Raster also in mono (zenith view). It is off by default, and the Raster button on the bar and this tick box are the same setting. The choices on the bar are saved with the program settings.

10.2 GND-Z and ground DTM

GND-Z is a modifier of AUTO-Z: turn AUTO-Z on first (Ctrl+Shift+Z), then enable GND-Z from the Snap or Cloud menu, the Object Snap toolbar, or Ctrl+Shift+Alt+G. Near zenith (view looking down), elevation prefers ASPRS ground-like classes (2 Ground, 8 Model key-point, 11 Road surface) when present in the neighbourhood; otherwise the lowest Z in about 1 m XY. Steep / façade views keep the normal RANSAC plane∩ray path.

Build the terrain model (CSF) (Terrain menu or Ctrl+Alt+B) runs a Cloth Simulation Filter on the Cloud AOI (preferred) or the current view extents. A progress dialog shows extract and CSF stages; Cancel discards the run. The resulting height raster is kept for the session: when GND-Z + AUTO-Z are on and the cursor is inside DTM coverage, elevation samples the raster (bilinear) instead of a live neighbourhood query. Discard the session terrain in the same menu drops it, and so does closing the project. A terrain built by the classification chain and written next to the layer is loaded instead with Terrain → Load the project terrain: each terrain on disk is listed with its size, and the one you pick becomes the session terrain that GND-Z, the contour lines and the volumes all sample.

10.3 WIRE-Z — collimating on a suspended line

WIRE-Z is the opposite reading of the same cursor to GND-Z, and like it a modifier of AUTO-Z: turn AUTO-Z on first, then enable WIRE-Z from the Draw → Snap, the Object Snap toolbar (beside GND-Z), or with Ctrl+Shift+Alt+H. The two are exclusive — arming one puts the other down.

With WIRE-Z on, the digitizing elevation comes from the surface nearest the camera under the cursor instead of from the strongest plane in the neighbourhood. It applies to AUTO-Z and to Cloud Z alike, so a click cannot take one reading while the hovering cursor showed another.

Why it exists. The zenith view is the one that shows a whole overhead line at once, so it is the view this work is done from — and from directly above, the strongest plane under the cursor is always the road: a conductor is a handful of returns against thousands of asphalt ones. Pointing at a wire and getting the ground eight metres below it is a property of plane fitting, not a setting that can be tuned. A suspended object has no plane to find; what identifies it is that nothing lies between it and the eye.

It is not limited to power lines. Anything that hangs or overhangs — a cable, a gutter, an eave, the near face of a structure from a terrestrial station — is what WIRE-Z lands on, in aerial, mobile and terrestrial projects alike.

Armed automatically. Choosing the overhead line category in the extraction panel turns AUTO-Z and WIRE-Z on and puts GND-Z down, and says so in the status line. It is not put back when the tool is released: a snap that switched itself off again would be one you could not rely on across tools. The Snap submenu always shows which is armed.

10.4 Drawing an overhead line

Cloud → Draw overhead line (Ctrl+Shift+Alt+A) arms a two-click tool: click twice on the wire, along it. The two points are not a convenience — they state which conductor and which way it runs, and near a bundle no measurement can recover that from one click. The span is then followed by contiguity from returns to returns, fitted, and drawn on the conductor layer with its length, sag, apparent section and fit rms reported in the log.

Arming it also arms WIRE-Z, so the cursor lands on the wire instead of the ground under it. The same menu item disarms the tool; WIRE-Z stays as you left it.

It needs a point cloud and a drawing open, and nothing else — no mobile run, and no particular panel. Choose the view that shows what you are tracing: the zenith shows a whole line at once, which is why it is the usual one.

10.5 Terrain — contour lines, volumes

The Terrain menu collects what concerns the ground surface, in the order the work is actually done. It needs a project open, and each group below needs the one above it:

None of these commands has a default key. All of them are in CAD → Keyboard Shortcuts… and can be bound there. GND-Z stays where it always was, under Snap: it is a snap, not a terrain function.

10.5.1 Contour lines

Terrain → Contour lines… traces the levels on the session terrain model and writes them into the drawing as polylines. The dialog holds what the drawing needs:

Undo the contour lines removes from the drawing exactly the lines the last run wrote. It is not a CAD undo: it erases those entities and nothing else.

If the dialog warns that the terrain is sparse, the raster has a height in too few cells: a contour segment needs four neighbouring cells with a height. Build the terrain on a tighter AOI, or with a coarser cell, before blaming the interval.

Breaklines are not available in this version. The automatic computation did not give reliable results and has been removed from the menus until it is studied further.

10.5.2 Volumes against a quoted polygon

Terrain → Volumes against a quoted polygon… (or type VOLUMI) measures cut and fill between the terrain model and an area you draw and quote yourself. It needs a terrain model and an open drawing.

With a project open, the same command is in the menu of the Terrain activity, in its last group with Volume strip and Cross sections.

1. Draw the area… — click its corners on the cloud; right-click or Enter closes it, Esc cancels it. The panel counts the vertices and the plan area as you go, and the last vertex can be taken back. A closed polyline already in the drawing can be adopted instead with Take the selected polyline.
2. Quote it. Every vertex arrives with the ground elevation sampled under it. Write the design Z over it in the table — vertex by vertex, or one value applied to all for a flat platform.
3. Compute. The calculation cell is yours to choose; the terrain's own cell is shown beside it, because a finer cell gives longer numbers, not truer ones.

The reference surface is the triangulation of your polygon, so every quoted vertex is honoured exactly. A least-squares plane through the same points would have thrown away the numbers the operator put there, which are the whole content of the measurement.

Two things are stated rather than hidden. Ground inside the area with no terrain underneath is reported in square metres and left out of the volumes instead of being guessed. And the share of the volume that rests on reconstructed terrain cells is given separately, with the part outside the surveyed area: a cubic metre over extrapolated ground is not worth one over measured ground.

Profiles brings the volume profiles forward in the bottom panel (see The bottom panel), optionally following the cursor, and the coloured blanket over the view shows where the difference is and how deep, at an opacity you set. Close command puts the drawn area, the blanket and the strip away without losing anything.

10.5.3 Cross sections along an axis

Terrain → Cross sections… cuts the terrain model square to an axis you have drawn, at a regular step, and gives you the sections as CAD polylines and as a table, with the longitudinal profile along the axis. It has no default key (assignable in CAD → Keyboard Shortcuts…).

What it needs

• The axis — one open polyline on the layer ASSE. The dialog tells you what it found: 1 polyline on the layer ASSE, N m, or that there is none, more than one, a closed one, or one with fewer than two distinct vertices. An axis that turns back on itself cannot carry sections.
• A terrain model — build or load the ground DTM first. If it is still loading, the dialog says to wait. Nothing is drawn or written without terrain, and there is no fallback.
• A project with its drawing open.

The dialog

• Step between sections (m) — distance along the axis between two sections. Sections fall on multiples of the step, counted from chainage 0 at the first vertex of the axis. Minimum 0.01 m.
• Width to the left (m) and Width to the right (m) — how far each section extends from the axis, measured on the ground plan, looking along the direction of the chainages. They cannot be negative or both zero. The default comes from the corridor of the job type.

The dialog warns in orange that the layer SEZIONI is replaced: everything on it is removed and redrawn, so keep nothing of your own there.

What you get

• The sections as 3D polylines on the layer SEZIONI, at their real elevation from the terrain. A section is split where the terrain has no value.
• The longitudinal profile on the ground-profiles layer, replacing a profile with the same ends.
• The table sezioni.csv and sezioni.xlsx in the project folder: one row per sample with section number, chainage, offset from the axis (positive to the right), E, N, ground elevation, and the origin of that elevation: M measured, I interpolated, E extrapolated, ? unknown, - none.

A progress cover shows the section being cut (Cross section n of m), then the writing of the drawing and of the table. On completion the status line gives the number of sections and polylines.

11. Cloud Performance

For very large clouds you can trade visual crispness against navigation smoothness with a single setting: Cloud → Performance. The choice is global and remembered between sessions. The settled LOD (once the camera stops) is intentionally non-linear: close to the subject it becomes dense quickly for restitution, while distant overviews are kept much coarser because they are navigation views, not drawing views.

What happens while the camera is moving (3.89) now differs by preset instead of being the same rule for all three. On Fast and Balanced the cloud is drawn at a coarser level of detail during pan, zoom and orbit, then one crisp full-resolution pass renders a fraction of a second after you stop — the momentary drop in density is what keeps the frame rate up while you move. On Quality the cloud instead stays at full resolution through the whole motion, orbit included: nothing degrades and nothing has to catch up once you stop, at the cost of doing the full-density work on every frame while navigating.

The Cloud menu also displays the active preset's live values (visible tiles per layer and motion level-of-detail bias) and current cache statistics (resident tiles, cache hits/misses).

Visual quality options in the same menu are independent from LOD. The surfel overlay increases point coverage, while Eye-Dome Lighting adds depth cues on the already rendered cloud texture. Both are saved in global settings and can be toggled without changing project geometry, CAD entities or pick behavior.

12. Stereoscopic viewing

The Stereo menu controls depth presentation. Stereo settings are saved with the project.

12.1 Quick start

1. Open a project with cloud (and CAD if needed).
2. Stereo → Stereo (or toolbar Stereo, Ctrl+Shift+Alt+S).
3. Stereo → Display layout — choose Mono, Stereo SBS (dual monitor), or PluraView (triple monitor), or cycle with Ctrl+Shift+Alt+H / toolbar Mono / SBS / PV.
4. Adjust Stereo → Adjustments → Depth & CAD quality if depth looks flat or lines too thin.

12.2 Menu layout

12.2.1 Selection and grips in stereo (3.50)

When you select a CAD entity in mono, it stays visible in external SBS with the same highlight colour and grips, at the correct stereo disparity, so you can edit in stereo. Customize selection (default magenta) and grip (default blue) colours in Inspector → CAD selection — mono and stereo share the same palette.

12.3 Display layouts

Monitors are labelled Display 1, Display 2, Display 3 in the UI — the same numbers Windows shows when you click Identify in Display settings.

PluraView triple monitor setup

With three monitors in Win+P Extend, choose display layout PluraView (triple monitor) (Stereo → Display layout, or Display setup wizard). Windows primary (Display 1) carries the UI and mono preview; each PluraView stereo panel gets one eye at full resolution.

1. Default — dual-window per eye: StereoCloud3D opens a borderless GL window on the left-eye monitor (e.g. Display 2) and another on the right-eye monitor (e.g. Display 3). No NVIDIA 3D Stereo driver setup required. Recommended first on RTX Quadro + PluraView.
2. Optional — quad-buffer passive: enable Stereo → Quad-buffer passive (NVIDIA clone). One stereo GL window uses OpenGL quad-buffer (BACK_LEFT / BACK_RIGHT). Requires NVIDIA Control Panel → 3D Stereo → On and passive stereo clone on the two PluraView outputs. The menu shows whether quad-buffer GL is available on the external window.
3. Assign eye monitors — use Display setup or Assign eye monitors… under the PluraView submenu when three displays are connected.

PluraView on two cloned monitors

On some benches the two PluraView panels are wired in NVIDIA Clone Mode, so Windows sees a single monitor rather than two. StereoCloud3D used to refuse that bench outright: the monitor count did not match what a PluraView was supposed to look like.

Turn on View → Stereo → PluraView options → Allow PluraView on two monitors (NVIDIA clone). It is off by default, because on an ordinary two-monitor desk it is not what you want. With it on, the application switches to passive quad-buffer and leaves the mirroring of the reflected eye to the driver.

The driver side has to be set once, in the NVIDIA Control Panel:

1. 3D Settings → Stereo — Display mode → Planar stereo mirror display. Needs an Ampere GPU or newer.
2. Set up stereoscopic 3D → enabled.

Mirroring the reflected eye in software

On a PluraView the eye seen through the mirror arrives reversed, and normally the monitor or the driver flips it. That leaves the bench configured for stereo even when you lift the mirror and want an ordinary 2D desktop.

View → Stereo → PluraView options → Mirror reflected eye (software) moves that flip into the application: the monitor stays neutral, and lifting the mirror gives you back a normal desktop with nothing to reconfigure.

12.4 Adjustments & cursor

• Swap left / right (pseudoscopy) — Ctrl+Shift+Alt+R or menu item under Adjustments.
• Stereo depth — Adjustments → Depth & CAD quality: normalized 0–1 slider 0 = flat, 1 = the maximum recommended relief. The whole travel is usable, so fine adjustments are easy. The value is applied when the slider is released (each change re-renders both eyes).

Stereo cursor & coordinatograph

In stereo the pointer is drawn as a dedicated reticle (ring or filled disc) that is perfectly stable and responds immediately to motion and elevation changes. It is pre-squeezed anamorphically, so through the stereo display it reads as a perfect circle, fused at the cursor's elevation. Default: filled red disc, 3 px. Size (0–10 px), thickness, shape (ring vs disc) and color are configured in Settings → Cursor → Stereo reticle. The standard CAD cursor (cross + pickbox) remains on the 2D editing view.

A coordinatograph (X, Y, Z readout) can be shown on the stereo output, per eye and anamorphic-corrected. The Z value is drawn at twice the X/Y size to ease setting the elevation. Enable it and set size/color in Settings → Cursor → Stereo coordinates. The readout uses the same floating-mark 3D point as the mono status bar and stereo reticle (3.20). During idle AUTO-Z, the readout is frozen on the sampled cloud point until the mouse really moves, avoiding planimetric jitter on stereoscopic panels.

12.5 Dual monitor (UI + external 3D panel)

Stereo SBS is routed to the secondary monitor; the primary display (Display 1) carries menus and a mono viewport. The recommended field setup is a laptop or FlexView (primary) plus a dedicated 3D panel (secondary). With layout Stereo SBS, StereoCloud3D enables the external window automatically when two displays are connected (Win+P Extend).

Windows display setup (required once)

Configure Windows before enabling dual-monitor mode in StereoCloud3D. The steps below apply to Windows 10 and Windows 11.

1. Extend the desktop — press Win+P (or Win+K on some laptops) and choose Extend / Estendi. Never Duplicate / Duplica: mirroring sends the same image to both screens and breaks the dual-window workflow.
2. Open display settings — Settings → System → Display (or right-click the desktop → Display settings).
3. Identify the monitors — click Identify. Windows shows a large number on each physical screen. Note which number is the laptop/FlexView and which is the 3D panel.
4. Arrange the monitor icons — drag the rectangles so their relative positions match the real desk layout (where you look left/right/up/down). Typical VoxelSpace setup: 3D panel on the left, laptop on the right. If your desk is different, match the icons to reality; you can pick the stereo monitor manually in StereoCloud3D (see below).
5. Set the main display — click the laptop / FlexView icon and enable Make this my main display / Rendi principale. The taskbar and StereoCloud3D UI should stay on this screen. If Windows primary is the 3D panel, the app still runs but logs a hint to swap primary back to the laptop.
6. Resolution — select each monitor and set its native resolution (or the project standard, e.g. 1280×720). Different resolutions per monitor are supported; the 3D view is scaled to the external panel automatically.
7. Scale (DPI) — under Scale & layout, prefer 100 % on the 3D panel when possible. High scaling (125 %, 150 %) on one monitor only can shift window edges slightly; if the stereo window does not fill the panel, check this setting.
8. 3D panel hardware mode — on PluraView / VoxelSpace, turn 3D ON on the monitor (hardware side-by-side). StereoCloud3D sends a side-by-side frame; the panel must be in 3D mode for depth to fuse correctly.

StereoCloud3D menu

• Stereo view mode ON.
• Display layout = Stereo SBS (dual monitor) for VoxelSpace / dual-panel setups, or PluraView (triple monitor) when three displays are connected.
• Dual monitor → 3D output monitor — which physical display gets the 3D-only window:
  • Auto (leftmost) — default for VoxelSpace-left / laptop-right layouts.
  • Auto (rightmost) — use when the 3D panel is the rightmost screen.
  • 0: … / 1: … — pick a specific monitor by name if auto-detection does not match your desk (listed in the same submenu).

Verify in the Command Line log

When dual-monitor mode starts, the Command Line panel prints a short Windows checklist and a line per monitor, for example:

Monitor 0: VoxelSpace @ (0,0) 1280x720 [stereo] (Win primary)
Monitor 1: FlexView @ (1280,0) 1920x1080 [UI]

Check that one monitor is tagged [UI] (your editing screen) and one [stereo] (the 3D panel). Tags may combine if misconfigured. Use Stereo → Dual monitor → 3D output monitor to correct the assignment without changing Windows layout.

For PluraView workstations, use Stereo → Verify PluraView setup... to audit the saved monitor IDs and the current external eye-window placement before a digitizing session.

Troubleshooting

12.6 Oblique views on oriented surfaces

Oblique views are the workflow for digitizing CAD features on surfaces that are not horizontal plan views: vertical façades, bridge abutments and pier faces, dam fronts, embankment slopes, roof pitches, retaining walls, and any other plane that cuts through a detailed 3D point cloud. Instead of forcing everything into a top-down WCS view — where steep geometry collapses into a thin line and depth is hard to judge — you tilt the view onto the real surface with pivot orbit and trackball rotation, then draw in true stereoscopic 3D with the same precision you already use on cartographic terrain.

Two view modes

Entities are always stored in World Coordinates (WCS) in the project file. An oblique view only changes how you look and digitize: the cloud and the CAD scene are tilted together to your work plane, and every click is converted back to WCS automatically.

Orienting the view

Load a cloud first, then orient the view from the cloud surface:

• Start a right-button drag over the structure: the pivot lands on it.
• Use Shift+MMB drag for yaw/pitch, or MMB + trackball ball for responsive hands-on rotation without dragging the mouse.
• The cloud and CAD scene rotate together during the gesture, so what you see is always what you will digitize.

Return to the ordinary plan at any time with View → Top / WCS or T: this restores the top-down WCS view and fits the extents.

12.7 UCS from cloud plane (RANSAC)

For façades and other large planar surfaces, you can fit the wall plane directly from the point cloud and lock an oblique UCS in one step — faster and more accurate than manual orientation tuning alone.

• Open View → UCS from cloud plane… or press Ctrl+Shift+U (with a cloud loaded).
• Pick a point on the face, set R (capture radius in metres) for the local RANSAC sample, then Apply. R affects only the plane fit, not what you see after clipping.
• The view aligns to the fitted plane: X horizontal on the wall, Z toward the camera.
• Clip (metres) controls a thin slab along the plane normal: only points on the visible wall within that thickness remain. The lateral extent follows the viewport (everything on screen on that face), not the RANSAC radius.
• Reset clip shows the full cloud again while keeping the oblique UCS.

Working on the oriented plane

Once the view is tilted onto the surface:

• The camera looks along the surface normal, so the wall/roof fills the viewport instead of collapsing into a line.
• The stereo cursor fuses at the elevation you set; parallax zero is at the cursor depth so entities sit on the surface you see.
• Mouse wheel (without modifiers) zooms the view. With Shift or Ctrl held, the wheel moves the cursor in depth — along the line of sight — which is the primary way to “lift” the cursor onto a façade or roof before placing a vertex.
• Cloud snap (Z) searches through the point under the cursor, so the picked elevation matches the cloud surface on that oriented plane.
• The coordinatograph shows the live WCS (X Y Z) coordinate of the cursor for export or reporting.
• A orientation triad (WCS axes) in the corner of the stereo viewport helps you stay oriented on oblique views.

Recommended workflow (façade / structure example)

1. Import the TLS or aerial cloud and open the project in Stereo (SBS on a 3D panel or PluraView triple for full-resolution eyes).
2. Pan/zoom to the structure; drag the horizontal slider to face the façade and the vertical slider to tilt the view onto the wall (or use the trackball).
3. Snap the cursor onto the cloud (Z), fine-tune depth with the wheel + Shift/Ctrl, then start a polyline or line macro and digitize the feature in stereo along the surface.
4. Press T (Top / WCS) to return to plan for spot heights or map-wide editing; your 3D vectors remain in WCS.

Typical applications

12.7 Fusion comfort on deep views

Stereo depth is produced by shifting the two eye images horizontally in proportion to depth. On an aerial nadir view the depth range is only the terrain relief, so that shift stays tiny and the whole model sits comfortably at screen depth. On a view with a large depth range — a road centreline seen from near ground level, a valley looking along its axis, a façade viewed edge-on — the far part of the scene can demand a shift larger than the distance between your eyes. At that point the image pair cannot be fused at all: a straight feature running away from the viewer appears as two lines opening into a V.

StereoCloud3D measures the depth actually visible in the current view and, when needed, reduces the stereo depth just enough to keep the whole picture fusable. The limit is computed from the physical geometry of your monitor — panel size and viewing distance are read from the display itself — against a comfort limit expressed as an angle at the eye: the largest separation allowed between the images of the two eyes. The default is 1.0 degree.

The limit engages by itself on deep views and lifts by itself when you zoom or pan back to a shallower one. When it is engaged, Stereo → Adjustments → Depth & CAD quality shows Limited to N% for fusion comfort under the depth slider, so a slider that appears not to respond is explained rather than mysterious. The Command Line log records each transition.

The Fusion comfort field

If zoomed-in views tire your eyes, lower the angle. In Stereo → Adjustments → Depth & CAD quality, under Stereo depth, the Fusion comfort field holds that angle in degrees. Type a value, or use the arrows of the field, between 0.1 and 3.0; the default is 1.0. A smaller angle allows less separation between the two eyes. When you are zoomed in close there is less depth to show and the overall view changes little, so lowering the angle costs little relief there; at low zoom the limit usually does not engage at all, and the field changes nothing you can see.

The value is a setting of your workstation, kept with the program settings rather than with the project: changing it does not mark the project as modified, and it applies to every project you open on this computer. The field is available while stereo is on.

The zero-parallax plane — the depth that appears exactly at screen level — is placed on the cloud surface at the centre of the view, so what you are looking at sits at screen depth and the rest of the scene spreads in front of and behind it. Moving the mouse never changes it: only panning, zooming or orbiting does.

12.8 Navigation view (looking only)

Measuring and navigating want opposite things from the same scene, and one setting cannot serve both. To read a few centimetres of relief on flat ground you need a large stereo depth; on a view that runs to the horizon that same setting pushes the far field hundreds of pixels apart and the pair stops fusing. Reduce it until the horizon is comfortable and a two metre tree in the foreground is worth a fraction of a pixel — no relief at all. The conflict is in the depth law itself, not in its tuning.

The Navigation view replaces that law with the one your eyes use. Depth comes from a perspective pair whose separation is derived from your monitor's geometry, so the far field stops at the comfort limit however deep the view is, while the near field keeps real relief. It is meant for flying over the ground and inspecting the model — not for measuring.

Turning Navigation view on with a saved camera path places the camera on the first frame of that path at once (paused). Play starts from there. Turning the view off restores the measuring camera.

Pan, zoom and orbit work exactly as before; zooming becomes a dolly, so the view moves toward what you are looking at instead of magnifying it. Zero parallax sits at the near edge of what is visible, which puts the whole scene at or behind the screen — nothing floats in front of the frame, the most tiring part of a grazing stereo view.

The mode is per session: StereoCloud3D always starts in the measuring view, so a restart is a guaranteed way back regardless of what happened before.

While the mode is on, an orange badge in the top right corner reads NAVIGATION — VIEWING ONLY, NO RESTITUTION. The view behaves like the measuring one under the hand, so the badge is what tells you why a drawing command is being refused.

12.9 Choosing your stereo desk layout

The Stereo menu lists every desk layout StereoCloud3D supports:

Layouts this computer cannot run right now are greyed out rather than hidden, and hovering one states what is missing — a second or third monitor, an emitter that is not connected. Plugging in what it asks for is enough; nothing has to be reinstalled.

12.10 NVIDIA 3D Vision 2 shutter glasses

If you own NVIDIA 3D Vision 2 glasses and their infrared emitter, StereoCloud3D can use them: each eye receives its own image at full resolution while the glasses close one lens at a time in step with the display.

Two paths, chosen automatically

How the two images reach the glasses depends on the graphics card, and StereoCloud3D decides by itself which of the two routes to take.

On a workstation card — a Quadro or professional RTX, with Stereoscopic 3D switched on in the NVIDIA control panel — the driver still produces both eyes, and StereoCloud3D simply hands them to it. This is the same route StereoCAD takes on the same computer. Nothing has to be prepared: no Zadig, no firmware file, no administrator run. In particular the refresh rate of the stereo display does not matter here, because the driver is what decides how the two images reach the panel: a passive polarized screen, a pair of screens in clone mode or a frame-packed 3D mode all show both eyes without alternating frames, and work correctly at 60 Hz and below.

On a consumer card NVIDIA's stereo has been gone since 2019 — driver 425.31 was the last one to carry it — so StereoCloud3D does the work itself: it loads the emitter's firmware, drives it over USB and alternates the eyes from a thread of its own. Nothing from NVIDIA's stereo software is needed or used. In exchange, the emitter has to be prepared once on each computer, and that path is what the rest of this section describes.

What you need, when StereoCloud3D drives the emitter

• Two displays. The interface stays on the Windows primary; the stereo output takes the other one over completely. This one applies to both paths.
• The stereo display at 120 Hz (100 Hz minimum). At 60 Hz each eye would receive 30 images per second and the flicker would be unbearable, so this path refuses to start. It is not asked of a workstation card, where the driver produces both eyes.
• The 3D Vision 2 emitter connected by USB, and the glasses charged.

A project in mono. Stereo is saved with each project. When the project you open is in mono, the 3D Vision display shows the mono view, the same in both eyes — flat but current, with the proportions of the mono view — instead of the last stereo image of the project you had open before. Switch stereo on from the Stereo menu and the project remembers it.

Start here: Stereo → Active shutter: what is missing

Before following any of the steps below by hand, open Stereo → Active shutter: what is missing. It measures this computer and lists, one line each, everything the glasses need: the two displays, which graphics card is drawing the window, the stereo panel and its refresh rate, which of the two paths the desk is on, and the state of the emitter. Green lines are done; the others say what to do.

Where the application can do the work itself there is a button rather than an instruction — setting the stereo panel to 120 Hz, opening the Windows display settings, restarting with administrator rights. Two things it cannot do for you, because they need your consent: installing the WinUSB driver for the emitter, and supplying an NVIDIA driver file when none is on the machine. For those it tells you exactly what is missing instead of leaving it to be guessed.

Preparing the emitter, once per computer

Windows has no driver for the emitter any more, so you install a generic USB driver for it. This is done once; it survives restarts.

1. Connect the emitter to a USB 2.0 port on the back of the computer — not a hub, not a front panel.
2. Download Zadig from zadig.akeo.ie and run it.
3. Choose Options → List All Devices, then select NVIDIA stereo controller in the list.
4. Set the driver on the right to WinUSB and press Replace Driver.

The emitter’s firmware

The emitter holds no program of its own: one has to be loaded into it every time the computer is switched on. That firmware belongs to NVIDIA and cannot be distributed with StereoCloud3D, so the application takes it from an NVIDIA driver already on your machine — the installed driver, the Windows driver store, or a file you place beside the application — and keeps a copy in %APPDATA%\StereoCloud3D so it only has to do this once.

On most machines this happens silently and there is nothing to do. If no such driver is found, StereoCloud3D says so and names the folder to use. Download an NVIDIA driver package old enough to still contain 3D Vision — 425.31 is the last one — and copy nvstusb.sys (or nvstusb64.sys) out of it into %APPDATA%\StereoCloud3D. You do not have to install that driver: the file alone is enough.

First run

The emitter keeps its firmware in volatile memory, so it is loaded again every time the computer is switched on. Completing that load requires Windows to re-read the device, which needs administrator rights. Start StereoCloud3D as Administrator (right-click its icon → Run as administrator) the first time after switching the computer on or after unplugging the emitter.

Then choose Stereo → NVIDIA 3D Vision 2, or press Ctrl+Shift+Alt+V. The stereo display goes full-screen and the glasses start shuttering. If they stay dark, StereoCloud3D shows a panel naming the step that is missing and what to do about it.

Adjusting the shutter

The moment the glasses switch relative to the image is calibrated for a 120 Hz panel and normally needs no attention. On a display that shows ghosting — a faint copy of one eye's image visible to the other — the delay can be changed with shutterIrDelayUs in the settings file, in microseconds. Close StereoCloud3D before editing it.

12.11 Vertical exaggeration

On gentle terrain the relief can be too shallow to read in stereo: a slope of a few metres over a kilometre gives almost no depth to fuse. Vertical exaggeration stretches the elevations on screen so the shape becomes legible, without touching the data.

You reach it from Stereo → Z exaggeration (view), from the Scene Tree context menu on the cloud, or with Ctrl+Shift+X, which cycles 1x, 2x and 3x. The menu also offers 1.5x and 5x and a slider up to 8x. Off is 1.0x.

When it is available. Vertical exaggeration is offered on an aerial project opened in stereo: elsewhere a stretched cloud would misrepresent the geometry you draw against, so the command stays inert. When it is unavailable it names the condition holding it back, so you do not have to guess which one applies.

It is also suspended while the Point section panel is open, because it reads depth from the picture and would report a wrong elevation while the stretch is on. Ask for the exaggeration with it open and the program offers to close it: confirm and the panel closes and the stretch is applied, cancel and nothing changes. Closing Point section loses nothing — every correction is handed to the drawing as you make it.

The stretch is display only. Measurements, coordinates, the elevation readout and the saved drawing are unaffected: what changes is how the cloud and the draped CAD are painted, so a 2x view is a picture of the terrain, not a new version of it. The setting is remembered between sessions, so if a view looks unexpectedly steep, check here first.

12.12 Professional stereo (fixed mark)

In ordinary stereo the measuring mark floats in space and follows your cursor across the stereo image: you move the mark to the model. Professional stereo works the other way round, as on analogue stereo plotters: the mark stays at the centre of the stereo window and in focus, at screen depth, and the model moves under it — sideways and in depth — while you move the mouse. You settle the mark on a point by bringing the point to the mark.

Turn it on with Stereo → Professional stereo (fixed mark) → Professional stereo (fixed mark) or with Ctrl+Alt+F. Both do exactly the same thing, and the choice is remembered between sessions. The status line confirms it.

• The mono cursor stays free. The mark is fixed only in the stereo window. In the mono view the cursor moves as it always does, and snap, automatic elevation, clicks and the elastic line of the CAD tools all work on it as before.
• Everything slides together. Cloud, CAD drawing and the cursor overlays of each eye move as one under the mark, so they keep agreeing in depth in both eyes. The three depth references described in this chapter are not changed.
• On every stereo output. It works with side-by-side and top/bottom stereo on an external display, NVIDIA 3D Vision, PluraView, quad-buffer, and with stereo shown in the main window.
• Georeferenced mono view only. The command is available once a project with its cloud is open and the mono view is the georeferenced one; elsewhere the program says so instead of doing nothing.

Why the stereo window stays full

As the model slides under the mark, the stereo window would run out of picture at one edge. To prevent this, the stereo images are redrawn centred on the cursor as soon as the mouse stops for about 0.2 seconds; and, while you are still moving, sooner (at most four times a second) when the slide has reached about a third of the shorter side of the window. Only the stereo images are redrawn: the mono view does not move. The part of the cloud that the mono view does not show is loaded for the stereo images as needed, so what you see there is cloud, not black.

12.13 Two viewpoints (top / bottom)

With Professional stereo on, the stereo window can be divided into two halves, one above the other, each a complete stereo picture with the mark at its centre:

• Top — the view of the mono window, exactly as in the single-viewpoint mode.
• Bottom — the same scene, at the same scale, turned about the centre of the mono view by a yaw and a pitch that you choose. Yaw turns it round the vertical; pitch tilts it. You see the point under the mark from two directions at once — for example the plan above and a side look below.

The menu is a submenu of the fixed mark: Stereo → Professional stereo (fixed mark) → Two viewpoints (top / bottom). It is available only while stereo is on and the fixed mark is on: without the fixed mark there is no split. Once it is ticked, two sliders appear in the same submenu:

Drag a slider and the lower half follows; Ctrl+click on it to type an exact value. You can also use the arrow keys, while the split is on and the program is in front: Left / Right change the yaw and Up / Down the pitch, by 5 degrees a press, or 1 degree with Shift; held down, they repeat. While the split is on the arrow keys go to this and no longer to the CAD. The status line shows the angles you reach. Every change is saved at once: the angles are the same the next time you open the program.

• The lower half carries the drawing too. CAD entities are drawn in the lower half, in each eye, and so is the elastic line of the tool you are using: its last segment ends on the mark.
• Both halves are filled with cloud. A half is much wider than tall; the picture is scaled to cover it (cropped above and below) instead of being left small in the middle with black around it.
• On every output of the previous section. With stereo in the main window the two halves are two panes with the proportions of the mono view.
• The yaw and pitch pivot on the centre of the mono view; the scale is the mono one.

13. CAD digitizing

The Draw and Modify menus (and the CAD Draw / CAD Modify toolbars) provide a familiar CAD command set. Combine them with object snaps and the cursor-Z controls to place geometry accurately on the cloud surface.

13.1 Drawing tools

13.2 Modifying entities

The Modify menu provides Move, Copy, Rotate, Scale, Mirror, Offset, Trim, Extend, Fillet, Stretch, Join (merge contiguous lines/arcs into one polyline), Square, Explode and Erase, plus Undo (Ctrl+Z), Redo (Ctrl+Y) and Cancel current (Esc).

Squaring an outline

Square (Modify → Square, the Sqr button on the Edit toolbar, the command SQ, or the right-click menu) makes drawn outlines orthogonal. Like every modify command it works on what you have already selected: select the closed polylines first, then run it. Several at once is fine. With nothing selected it refuses and says so — it never waits for a further click.

The reference side is the longest one, and the command picks it for you. That side keeps its direction; every other side that is already nearly square to it is brought to exactly 90°. The longest side is the best measured one: on a ten-metre wall a ten-centimetre error is half a degree, on a one-metre wall it is five.

A side that is genuinely oblique — a wall that really does run at 60° — is left where it is, together with its neighbours' corners. Squaring straightens what you drew slightly crooked; it does not invent a rectangle where the building is not one. Re-entrant corners (the inside of an L) are squared to 270° exactly like the outside ones.

Three tolerances govern it, in the [Squaring] section of %USERPROFILE%\StereoCloud3D.ini: AngleTolerance (default 10 degrees) is how far from square a corner may be and still be corrected; ShiftTolerance (default 10) is the furthest a vertex may travel — beyond it the correction is refused rather than applied; OffsetTolerance (default 1) covers the case of two parallel sides meeting. Raise the angle tolerance for outlines traced by hand on a sparse cloud.

The polyline is edited in place: it keeps its id, its layer and its selection, and one undo step puts it back.

Undoing an edit the engine never saw

Ctrl+Z is the CAD engine's undo, and it covers the engine's own commands. The corrections the application makes itself do not go through one: dragging a pitch handle, snapping a vertex or an edge with Fix vertex, deleting a building's pitches. For the engine those edits never happened, so Ctrl+Z after a handle drag would not undo the drag — it would undo the previous CAD command, which is a different kind of damage, not a remedy.

Moving a vertex with its grip is one of them: the engine writes the vertex but does not record it. Since 5.0.2 the program records it for you: Ctrl+Z (and Modify → Undo) puts the vertex back as long as no CAD command was run after the move. Releasing the grip without moving it, or with Esc, leaves nothing to undo. It covers polylines, up to 200 selected entities.

Modify → Undo geometry edit (Ctrl+Shift+Z) is the stack that was missing, and the menu entry names the gesture it is about to undo. Two stacks rather than one are deliberate: merging them would mean guessing which of the two histories you meant, exactly while you are repairing a mistake.

13.3 Entity properties

Open the Entity Properties panel from CAD → Entity Properties…, the Ctrl+1 shortcut, the PROPS command, or by right-clicking an entity. The panel follows your selection and lets you inspect and edit attributes such as layer and color. The vertices of a polyline or line are listed and edited in the Vertices dialog.

In inspection mode (no draw command active), select a line or polyline to show grips: green for the start vertex, blue for the others. Drag a grip to move that vertex in X/Y/Z like StereoCAD.

Vertices dialog (5.8)

One dialog edits the vertices of the selected polyline or line. Open it from the Vertices group of the left toolbar, from the right-click menu → Vertices… or with the VERT command. It stays open and follows the selection; a left-click in the drawing is still a plain CAD selection.

• Summary — number of vertices, length in plan and in 3D, elevation range and, for a closed polyline, the area in plan.
• Vertex list — numbered as in the drawing, with X, Y, Z, side and slope to the next vertex. Click selects one vertex, Ctrl+click adds or removes one, Shift+click selects a range; All, None and Invert do the rest. Selected vertices have a light blue ring in the drawing. Type a coordinate in a row and press Enter to move that vertex.
• Elevation — on the selected vertices (or the current one): Assign Z gives them one elevation, Shift Z adds a difference, Interpolate Z puts the vertices between the first and the last selected on the straight line between those two. X and Y do not change.
• Vertices — Delete the selected ones (a closed polyline keeps at least three, an open one two), Insert after the current vertex (halfway along the next side), reverse the direction, and Copy coordinates (number, X, Y, Z) to the clipboard.

Every change can be undone with Modify → Undo geometry edit. The vertex list that used to be in Entity properties and the old Vertices window are replaced by this dialog.

Add and Remove live where they act: right-click on an existing vertex of the selected polyline and the small Add / Remove bubble opens on that corner. Add inserts a new vertex halfway to the next one (already on the outline, ready to be dragged); Remove deletes the one clicked. Clicking elsewhere or pressing Esc puts the bubble away. A right-click that does not land on a vertex opens the ordinary context menu.

Dragging a vertex obeys the same rules as drawing it. While the corner follows the pointer the object snap stays live: the marker is drawn under the cursor and a found endpoint, midpoint or intersection wins over the cursor quota, so a corner can be welded onto existing geometry. AUTO-Z (Ctrl+Shift+Z) and Cloud Z measure the elevation when the pointer stops, and the vertex takes that quota instead of keeping the one it started from. A wheel Z step still wins over both until the pointer moves again.

While you drag a grip, the entity being edited is not offered to the snap as a target: the vertex no longer snaps onto itself. Endpoint, midpoint, intersection and the other snaps look at the other entities (and a polyline can still close on its own first point).

For a visual check, enable View → Entity vertices or press Ctrl+Shift+Alt+.. StereoCloud3D draws cyan, read-only markers on CAD line, polyline and point vertices; the overlay is for cartographic inspection and does not write to the DWG.

13.4 Layers

• Layer Manager (CAD → Layer Manager…, command LA) — create layers, set color and visibility, and lock a layer to protect its entities from selection/editing. Click a layer’s color swatch to open the RGB color picker for a custom color (the ACI index is updated automatically; you can still type an ACI number directly). Select all / Deselect all switch the On column for every layer the Show filter is listing, so you can blank the drawing and bring back one layer at a time (with Used layers selected the empty layers are left untouched).
• Current layer (3.71) — new entities always go on the current layer. The hidden SC3D_CLOUD_BOUNDS layer (cloud limits used only for framing) is never made current. If layer 0 is missing, StereoCloud3D creates it white (ACI 7) with default settings and makes it current. The Layer combo and Make Current refuse the bounds layer.
• Layer Order (command LAO) — control the draw order of layers so important features render on top.

The Description column and the code library

The Layer Manager carries a Description column beside the name. It is not a field of the layer — the CAD engine has no such field — it is what the layer means, read from the code library.

The library exists because in survey practice the code is the primary thing and the layer follows from it: you pick a code and the layer, the colour and (in time) the symbol come with it. StereoCloud3D ships the Italian national catalogue as its base: the Catalogo dei Dati Territoriali of DM 10 November 2011 (Gazzetta Ufficiale n. 48, 27 February 2012) — 11 layers (Strati), 36 themes (Temi) and 112 classes, each with its mnemonic (EDIFC, AC_VEI) and its hierarchical numeric code (020102). Hovering a description shows both.

That catalogue was not chosen for its size. It is the one the Italian Regions derive their own codings from, which is why a code from Lazio and one from Sardinia — different strings — can be traced to the same class. Codes differ between authorities; the criterion underneath them does not have to.

The library is built in two levels:

• base — the standard, read-only, shipped with the application;
• client — your own package, which references the base codes by mnemonic and adds the layer names you actually use. It does not redefine the standard, so the standard stays one.

A client package ships with the application too, mapping the layers StereoCloud3D's own tools create — the pitch, eave, building-footprint and road layers — onto catalogue classes. The wall layer is left unmapped: the catalogue has no class for a wall surface, and the nearest one is a volume. It lists every language those layers can be named in, because generated layer names follow the interface language.

A layer that is in no package has no description, and that is deliberate. An empty cell tells the truth; a description guessed from a similar name does not. Generic geometry layers (polyline, spline, ellipse, rectangle) are left unmapped for the same reason: they carry no survey meaning.

Codes the standard does not have. A client package can also define its own codes, not only point at catalogue ones — the manhole type a particular contract insists on, a working reference your office uses. A defined code carries its own names and its own layer, and it can declare where it belongs: give it a Strato and a Tema and it appears in that tab, among the catalogue codes the operator would look through anyway. Leave that out and it lands in a Your codes tab of its own, rather than in a place we picked for it by resemblance.

You can see which is which. Codes from the standardised catalogue are drawn in gold; your own stay in the ordinary white of everything else. The direction is deliberate: what is worth marking is the outside provenance, not your own codes, which are the ones you use most.

Studio packages live in %APPDATA%\StereoCloud3D\codes\ — any .json there is loaded, in name order. That folder, not the program folder, is where they belong: an application update rewrites the program folder, and a library built over two years must not go with it. esempio_studio.json, shipped beside the application, is a template to copy there and edit.

Creating a layer from a code. Because the code decides the layer, the Layer Manager has a From code… button beside New layer. It opens the catalogue with one tab per Strato and a heading per Tema — the structure comes from the catalogue itself, so adding a package changes what the dialog shows without changing the dialog. Type in the search box and the tabs step aside for a flat list: someone looking for a manhole does not know which Strato to open, and that is exactly when search earns its place. Picking a code creates its layer if the drawing has none, and makes it current, so what you draw next lands there.

Typing a layer name by hand stays where it was, for the layers a catalogue does not cover.

The library is customer data, not project data: it travels from job to job and grows as you add codes, and a project refers to it rather than owning it.

13.5 Entity groups

Group any set of entities — regardless of their layers — so you can hide or lock them together. Select entities, then Group selection (Ctrl+G). Manage groups in the Entity Groups panel (Ctrl+Shift+G), where each group has Hide and Lock toggles, a Select action, and a delete button. Use Ungroup selection to dissolve a group.

13.6 Eave outline

Draw → Eave outline starts a closed polygon digitized directly on the mono view, one vertex per click, drawn already closed — the same criterion as the elastic polygon in Building reconstruction: from the second vertex on you see the region you are enclosing, not a path you have to imagine. Unlike an ordinary CAD polyline, the vertices are held by the application itself and are not written to the drawing until the outline is closed. This is what lets Multiview correct any of them — including the one you just placed — without the limitation that normally stops a vertex being moved while the drawing command that placed it is still open. This is a standalone tool for an eave outline that stands on its own, independent of any pitch; when a building is drawn from its pitches (13.7 onward) the eave outline is derived automatically and there is no button for it in Multiview.

• The elevation you see is the elevation you get — a vertex takes the elevation shown in the status bar and in the loupe, provided that reading belongs to the pixel under the cursor (within 3 px); otherwise a fresh sample is taken and the log says which of the two answered. There is no longer a second, independent sample at click time, and therefore nothing left to reconcile afterwards: if the number is wrong you see it before clicking, rather than having it quietly adjusted after.
• Elevation lock (Shift) — hold Shift while placing a vertex to skip the cloud sample entirely and use the previous vertex's elevation instead, at whatever plan position the cursor is over. Use it for a vertex hidden under a tree or other clutter, where the cloud gives no reliable surface at all — hysteresis only rescues a vertex the cloud sampled wrong, this bypasses the sample outright. A gold padlock appears next to the cursor whenever Shift is held and a previous vertex exists, showing the elevation it would lock to before you click.
• Layer — the outline is written to a layer assigned to the eave entity (default GRONDA), not to whichever layer happens to be current when you close it. Edit the field in Multiview; the setting is persistent, and the same layer is what a confirmed building's automatic eave outline (13.9) uses too.
• Closing — right-click, or click near the first vertex, closes the outline, writes it to the drawing on the assigned layer, and immediately starts a new one of the same type: no need to reopen the command between buildings.
• Cancel — re-selecting Draw → Eave outline, or Esc, while drawing discards the vertices placed so far without writing anything.

13.7 Pitch face

Draw → Pitch face (or the button inside Multiview) digitizes one roof pitch as a closed face (usually 3 or 4 vertices, but with no fixed cap — an L-shaped pitch or a broken gable needs more), exactly the same way as an eave outline — the elastic polygon lives in the application until it is closed, so Multiview can correct a vertex, including the one you just placed, while the command is still open. The only difference from the eave outline is the layer it is delivered to.

• Snap to an existing vertex — a pitch vertex prefers a nearby vertex already on the pitch or eave outline layer — including an earlier pitch's corner — so two adjacent pitches share the exact same corner instead of leaving a crack between them. A large green square with a solid dot marks the point a click would take, visible as soon as the cursor gets close, before you click. The snapped vertex carries the existing corner's exact elevation.
• Snap to the nearest point on an edge — when no vertex is near enough, the click snaps to the closest point on an existing edge, with the elevation interpolated along that edge. This is what lets a new pitch meet an existing one part-way along a ridge. A vertex is always preferred over an edge at equal distance, because it is the more precise target. A vertex landing in the middle of another pitch's edge makes a T-junction, which the building grouping welds for you — see 13.8.
• Free vertices onto the pitch plane — a pitch is a plane: the first three vertices define it, and from the fourth on the elevation is no longer new information but a consequence. With this checkbox on (default, in Multiview), a vertex you have not fixed yourself takes its elevation from that plane instead of from the cloud sample under the cursor. The plane starts from your own three vertices — it is what you declared, and it is not up for discussion — then only the cloud points within 50 cm of it are kept and the plane is re-fitted by least squares on those, thousands of points instead of three clicks. If the refined plane has tilted more than 15° from yours it is not a refinement and yours is kept, so the result can never run away. A vertex you fixed by snapping, by holding Shift, or with the wheel is never touched: those are decisions, not measurements.
• Elevation lock (Shift) — same mechanism as the eave outline: hold Shift to lock a new vertex's elevation to the previous one instead of sampling the cloud, for a vertex hidden under a tree. It only applies when the cursor is not already snapped to an existing pitch/eave vertex (above) — a snap already carries its own exact elevation, so Shift has no effect there.
• Layer — a layer of its own (default FALDA), separate from the eave outline's, edited next to the button in Multiview.
• Filled — optional checkbox in Multiview: on close, the face is delivered as a filled solid (a polyface mesh) instead of a plain closed outline, with a dark outline added on top so its shape reads even against a dull fill colour. The triangulation is ear clipping, which holds for any simple polygon including re-entrant ones: a fan cut from the first vertex is only valid on a convex polygon, and on an L-shaped pitch its triangles jumped the notch and the fill spilled past its own outline. The Semi-transparent 3D faces checkbox, in the same panel, dithers every filled face so the cloud and anything behind it stay visible through it.
• Closing — right-click, or click near the first vertex, closes the face early (from the 3rd vertex on), writes it to the drawing, and immediately starts a new one.
• Undo a vertex — Esc removes the last vertex, not the command: it is the gesture you want when a single click went wrong. Esc twice within 0.6 s abandons the whole face. The eave outline behaves the same way.
• Guard against impossible geometry — a face whose own bounding box spans more than 100 km is refused rather than written, naming the offending vertex. A single bad elevation used to be enough to make a drawing unusable.

Pitch handles — moving a corner or a whole edge

Every pitch already in the drawing carries handles, always present, with no command to start — the way grips work in a CAD. A filled light-blue dot sits on each vertex and a hollow ring on the midpoint of each edge. Hovering one enlarges it, turns it gold and turns the pointer into a hand; the grab radius is deliberately wider than the dot, because aiming is never pixel-perfect.

• A vertex handle moves that corner. An edge midpoint handle translates the whole edge rigidly, both ends together, so it keeps its length and its direction.
• Everything welded to that point moves with it. Every vertex within 3 cm of the one you grabbed — on the pitch, eave, foot and wall layers alike — moves by the same amount. Moving one entity on its own would unstick the building, which is exactly the defect the handles exist to remove.
• The movement is in plan. The cursor is intersected with the horizontal plane at the handle's own elevation, so the metres are real metres and not an invented scale. Elevation is not touched: for that there are the wheel and Multiview.
• The fill follows the border, because the CAD engine exposes a filled pitch's mesh as a polyline too, so it is moved by the same write.
• Handles disappear while a pitch or an eave outline is being drawn, so a click stays a new vertex and is never stolen. They are drawn behind windows, never over a dialog.
• On release the status bar reports how far the handle moved and how many entities moved with it.

13.8 Numbering and drafting a building

As soon as a closed pitch is delivered (13.7), the pitches on the pitch layer are grouped by shared edge — two pitches that meet along a whole ridge, hip or valley belong to the same building — and each group found is marked with a small numbered circle drawn directly on the mono view: no button, no menu item, it just happens. Two pitches that only touch at a single point (a hip apex) are not joined yet — only a shared edge counts, for now.

T-junctions are welded first. A vertex placed with the edge snap (13.7) lands in the middle of another pitch's edge: geometrically correct, but the two faces then share no whole edge at all, so they would not be grouped and their shared boundary would never cancel. Before any comparison, every vertex that lies on another pitch's edge (within 3 cm, measured in 3D) is inserted into that edge, so P→Q becomes P→M→Q and the two faces really do share M→Q. This is done on the reading, never on the drawing: a filled pitch is two entities carrying the same vertices, and inserting a vertex into the mesh would shift its face indices and tear the fill. What you drew stays what you drew.

A filled pitch counts once. The mesh and its dark border are both exposed as polylines with identical vertices; counting both made every filled pitch look like a two-pitch building whose outline could not close. The duplicate is dropped, and the log names the two entity ids that were paired. A pitch drawn later, anywhere in the survey, re-groups and can renumber every building (west-to-east, south-to-north order), but never forgets which ones were already confirmed (13.9) — that is tracked by which pitches a building owns, not by its number. A project opened with pitches already on the layer from an earlier session numbers them once Draw → Number buildings is run by hand, since there is no new pitch to trigger it automatically.

Clicking a circle selects that building (a gold ring appears around it) for the Confirm controls below; clicking empty space clears the selection. Circle colour says what state a building is in: cyan is a draft — nothing has been written to the drawing yet — and green is confirmed (13.9). Clicking a confirmed building's circle reopens it as a draft: its eave outline, foot and walls are deleted from the drawing and it goes back to being a live preview, editable again until re-confirmed.

A selected draft building previews its eave outline (gold) and its foot (orange) as outlines on the mono view, recomputed every frame from the current eave overhang setting (13.9) — neither is a CAD entity yet. The outline itself is the union, in plan, of the building's pitches: cancelling the edges shared between adjacent pitches (the same edge, walked in opposite directions by the two pitches that share it, present twice in the raw edge list and removed as a pair) and chaining what is left into one ring. A building whose pitches do not close a single ring — a gap, an unfinished pitch — has no preview, and says so on the mono view, next to its own number: an outline that is refused in the log only looks like a building that simply does not work. The refusal itself is deliberate. Grouping accepts a shared edge walked in either direction, while cancelling demands opposite directions; when fewer edges cancel than the group needs, the two tests are contradicting each other and the "outline" that comes out runs along the pitches one after another and passes through the middle of the building. Better no outline than a wrong one, because a wrong one becomes a foot and walls written into the drawing.

Every group is also filed as a product. Each becomes one entry under Products → Buildings in the project tree, named by its number, owning the pitches that make it up — drafts included, not only confirmed buildings. That is what makes a hand-drawn building countable, datable and switchable: its tick box hides its own pitches, and the tick box on the Buildings folder hides them all — drawn and machine-reconstructed alike, because both live in the same family with their origin recorded inside rather than in a separate branch. A re-grouping replaces these entries the same way it replaces the numbering; entries from the automatic building reconstruction are left untouched.

13.9 Confirming a building: eave outline, foot and walls

Changed in 5.2. The ground height of the foot is sampled 2 m outside every side of the outline, where the ground is visible, and only heights at least 1 m below the eave count (median of the samples). It used to be sampled at the centre, under the roof, which could put the foot above the eave and build the walls upwards. With no ground found the building is not confirmed and the status line says to build the terrain model. The Eave overhang is a slider from 0 to 2 m in steps of 10 cm.

The controls sit in Multiview, below Pitch face. Confirm writes three entities together for a building still in draft — the eave outline (the building's own outline, on the eave-outline layer, 13.6), the foot (that outline offset inward by the eave overhang, at the ground elevation under the building, on its own layer, default PIEDE) and the walls (one solid mesh per building: the eave soffit where there is an overhang, plus the wall itself down to the ground, on its own layer, default PARETI) — and marks the building confirmed. To redo a confirmed building, reopen it first by clicking its circle (13.8).

• Eave overhang — the field in the panel, in metres, persistent (default 0.50 m). The foot is the outline offset inward by this amount in plan, and the eave soffit is exactly this wide — it is the same construction for both.
• Ground elevation — sampled under the building's outline centroid the same way as GND-Z (the terrain raster first, the point cloud if there is none), falling back to the lowest outline vertex when neither answers.
• Foot and wall layers — each its own field in the panel; the eave outline uses the layer set in 13.6.
• One building at a time — select a building (click its circle) and Confirm building N writes that one. There is no "confirm all": a button that writes geometry into the drawing for buildings you are not looking at is a button that gets pressed by mistake.
• Wall handles — a selected draft building shows an orange handle on the midpoint of each wall of its foot. Dragging one changes the eave overhang for that wall alone, along the direction that wall actually moves in, so a movement parallel to the wall does nothing. It is how you push one wall up against the neighbouring building without changing the rest. The per-wall values are written when you confirm.

13.10 Road

Draw → Road digitizes a road as a set of parallel lines while you follow only one of them. You click one vertex per click along a single edge you can actually see in the model — the guide — and the other lines are carried along beside it. The right button delivers the road and starts a new one immediately; Esc removes the last vertex.

The companion lines are not copies. Only the guide takes its elevation from where you click. Every other line takes its own elevation from the cloud, sampled under its own position at the moment the road is delivered. That is the point of the tool: the result carries the real cross-slope of the carriageway instead of a flat ribbon offset from one edge. On a cambered road, or where the far edge rises into a kerb, the difference is the whole measurement.

Each line is delivered as an open polyline, all of them on the road layer, colour ByLayer.

The settings live in Multiview, Road tab:

• Number of lines — 1 to 3 (default 2). Two gives the pair of edges; three adds the axis or a second carriageway edge.
• Spacing — distance between adjacent lines in metres (default 6.00), the nominal carriageway width.
• Guide line — which of the lines is the one you are digitizing (default the first). Set it to the middle line to follow the axis and have both edges derived, or to the last to follow the far edge.
• Road layer — where the polylines are written (default STRADA).

Pick the guide that you can see: the edge with a kerb, a white line or a clear change of material. The tool is at its best when the line you follow is the one the model shows most sharply, and the ones you cannot see clearly are left to the cloud.

13.11 Building outlines

On an aerial survey the program can propose the outline of every building by itself. The proposals are ordinary closed CAD polylines, drawn on two provisional layers, and you judge them by eye: correct what is wrong, then accept what is right. Nothing is final until you accept it. This section covers how the outlines are made, how to show them, how to correct them, how to check them and how to send them to a GIS.

Both layers are drawn thick (0.50 mm) with their vertices marked, in mono and in each eye in stereo. An outline smaller than about 24 pixels on screen shows only its line: from far away the vertices would cover the whole cloud, and they come back as you zoom in. A courtyard of 25 m² or more is a separate closed polyline on the same layer as its building. The layer names are fixed and are not translated. Each vertex carries the elevation of the eave.

13.11.1 Showing the outlines, and when they are made

• The button. In aerial jobs a Building outlines button sits at the bottom right of the view. It is yellow while the outlines are shown. It needs an open CAD drawing: without one it is grey and its tooltip says so.
• First click. When the drawing holds no outlines yet, the click makes them from the cloud (see 13.11.2), with a wait window that shows the phase and, when it counts, “Building outline i of n”.
• Afterwards the button only shows and hides the two layers and keeps every edit you made. The same switch is the Building outlines (hypotheses) box in the Display panel and the entry of the same name in the View menu. It has no default key; assign one under Building outline hypotheses in Keyboard Shortcuts if you want it.
• Always off when a project opens. The first time a project's drawing is loaded the two layers are switched off, so the view is ready at once. You switch them on from the button, or from the Building outlines box at the start of the Buildings tab (the button can be hidden under the Display panel). The same switch shows and hides the 3D roofs (Roof 3D).
• Never computed when a project opens. Opening a project does no building search at all. If the outlines in the drawing were made by an older version of the recognition, a window stays on screen until you press OK and tells you which version made them and how to redo them (Automation → Regenerate building hypotheses or Automation → Building outlines preview). Nothing is regenerated for you. Outlines that were made with the AI (13.11.4) get the same notice, with the reason that redoing them uses AI credits.

13.11.2 How the outlines are made

The search runs on this computer, reading the first cloud layer of the project that has an octree. It does not change the cloud or its classes. In outline:

• Elevation raster at 10 cm. The highest surface of the cloud is rasterised at 10 cm cells, empty cells filled. The terrain comes from class 2. A cloud with no class 2 is still accepted: the program works out a terrain of its own for the search and leaves your classes untouched.
• Roofs. Continuous, high, flat surfaces are kept; tree crowns attached to a roof, small annexes and small holes are removed. Houses that touch are separated at steps taller than 0.4 m and at walls.
• Vegetation. On a classified cloud, wide patches whose highest point is vegetation (classes 3, 4 and 5) are never roof, and a surface with vegetation on top over more than half of it is dropped whole: dense crowns no longer come out as buildings. On an unclassified cloud with colours, a surface that is green on top and has almost no flat part is a crown and is dropped; green roofs stay, because they are flat.
• Shape and size. An outline must be a building someone could use. A piece narrower than 3 m at its widest point - a wall, a fence, a kerb - is dropped; a piece narrower than 6 m that is also lower than 3 m and green on top - a vegetable garden, a pergola, a pool cover - is dropped too. Sheds and garages of that size stay. A narrow strip as high as a roof that runs along a building for at least half its length - an eave or a lower pitch cut off by a step - is joined to that building.
• Sides. Walls seen obliquely are discarded, and each side is marked as seen, covered by a higher crown or not detected. Seen stretches of at least 2 m define the walls; the outline is squared on the dominant wall direction, and sides that were not seen are closed by prolonging the walls at right angles.
• Neighbours. Where two houses share a wall the two sides are put on the line of the higher roof; where outlines overlap, the overlap is taken away from the lower building.
• Density filter. The mean point density inside each outline is compared with the typical density of the whole survey. An outline whose density is below a quarter of that is removed: it is a false building on a thin strip of points, usually at the edge of a flight line.

LiDAR and photogrammetry

• LiDAR cloud. The edges are already sharp in the cloud, so the outlines come from the cloud alone. The program recognises LiDAR from the laser intensity of the points.
• Photogrammetric cloud with an orthophoto in the project. The roofs are drawn on the orthophoto by the program's internal AI (it runs on the computer, free, no AI credits); the cloud supplies the elevations. The wait window shows Roofs from the orthophoto (AI) with the tile count. Because photogrammetric clouds give noisy shapes, the following shape rules also apply to them: outlines under 8 m², strips narrower than 2 m, shapes that fill less than 55% of their minimum rectangle, self-crossing outlines and corners sharper than 30 degrees are removed, and a shape that fills 85% or more of its minimum rectangle becomes that rectangle.
• Photogrammetric cloud without an orthophoto. The edges of the cloud alone are often too soft. The program measures the sharpness of the edges (the share of the outline that is made of straight walls of at least 2 m); below 40% it creates nothing and says so in a window that gives the measured percentage and, for a photogrammetric cloud, tells you to add the orthophoto to the project.

13.11.3 After “Classify cloud”

On an aerial survey with a drawing open, the provisional outlines are made again at the end of Classify cloud, because the classification has changed: the old provisional ones are removed once the new ones are ready, and the status line adds the number of outlines created and the seconds taken to the classification summary. Outlines you have already accepted are on other layers and stay; inside them the search proposes nothing. A project with no provisional outlines gets none: only outlines that already exist are kept in step with the classes. With no drawing open the status line says that the outlines were not made. Outlines that were made with the AI are never redone this way: a notice says so, and you regenerate them from the menu when you want. A ground (mobile) survey does not run the search at all.

13.11.4 Preview on an area, and choosing the method

Automation → Building outlines preview… lets you see, on a small area, what each method gives and what it costs, before running the whole project. It needs a cloud with an octree and the project's orthophoto. It writes nothing to the drawing.

On a LiDAR cloud it is not offered (the entry is not in the menu, and the Process panel says why): the outlines come from the cloud and there is nothing to compare. To try a block first, draw an area of interest and run the automatic outlines.

1. The area of the preview is the area of interest. With an area set, the preview starts at once. Without one, the area window opens ready to draw: left click the vertices, right click to close, and the preview starts when the polygon is closed (Esc with no vertex cancels). A wait window follows.
2. A window shows the orthophoto of the area twice, with the outlines in yellow: Local and AI. Each column says how many buildings it found in the area.
   • Local is free and runs on this computer: on a LiDAR cloud it works from the cloud, on a photogrammetric cloud it takes the roofs from the orthophoto (13.11.2). It is the one marked recommended. The column also gives the time for the whole project.
   • AI sends the orthophoto of the area to the AI service and uses AI credits. The column gives the credits this preview used and an estimate for the whole project in credits and euros (the area is scaled to the project: dense areas cost more, open country less). The AI outlines are kept as the AI drew them, with the eave elevation taken from the cloud and overlaps removed from the lower building.
3. Use for the whole project under a column regenerates all the outlines with that method. If the drawing already holds provisional outlines you are asked to confirm, as in 13.11.5.

Which method made the outlines is remembered with the project. A project made with the AI is not redone by itself when the recognition is updated or after a classification (13.11.1, 13.11.3).

13.11.5 Regenerating the hypotheses

Automation → Regenerate building hypotheses… asks for confirmation (“The n provisional outlines, with your corrections, will be replaced by new ones computed from the cloud. Accepted buildings stay and are not recreated.”) and then replaces the provisional outlines. The old ones are removed only when the new run succeeds; if it fails or is refused for soft edges you find them as they were. Hypotheses that would fall inside an accepted building are not made again.

Confirming replaces the whole draft: every provisional outline and every 3D roof pitch, dormer and chimney (Roof 3D) is removed, because they belonged to the old outlines. The report says how many pitches went.

Only inside an area of interest. When an area of interest is set, the search reads only that area (with 20 m around it, so that a building across the border comes out whole) and keeps the buildings whose centre is inside the polygon. On a large project this is the quick way to work a block at a time: a few seconds instead of minutes. The old draft is still removed everywhere, also outside the area; the report and the status line say so.

During the run the progress bar follows the time actually spent, and after a minute it shows how long is left (“about 8 min left”), estimated from the work already done. At the end a report says how many outlines were created and in how long, what was not drawn because it is not a building (tree crowns, pieces narrower than 3 m, green gardens and pergolas, outlines on too few points) and what was corrected (eaves joined to their building, shared walls, overlaps).

Which method: the program chooses from the data

Clicking Building outlines in the graph opens a menu whose first line says what your data allow: LiDAR: outlines from the cloud; Photogrammetry + orthophoto: edges from the orthophoto; Photogrammetry without orthophoto: no automatic outlines (add the orthophoto or draw by hand). Below it, Automatic outlines - whole project and, indented, try first on a small area - the same choice, tried before. With an area of interest set the two entries read Automatic outlines - in the area of interest and compare the methods in the area of interest; on a LiDAR cloud the second one is not in the menu. Methods that cannot work with your data are greyed, with the reason. The hand tools come after, under Then correct or add by hand.

The Buildings tab: one outline at a time

When the drawing has provisional outlines, the bottom panel shows the Buildings tab: the list of the provisional outlines, grouped into Seen and Sides under trees, ordered by position so that the next one is close to the current one. Previous and Next frame and select an outline; the current one is drawn with a thick magenta border, in mono and in both stereo eyes.

The tab has three more rows.

Nothing lost in silence: the items to verify

What the search finds but does not draw as a building is not thrown away. Areas that are high and flat and not vegetation, but that no outline covers, and outlines dropped for sparse points or an improbable shape, are drawn as blue polylines on three layers of their own and listed in the Buildings tab, largest first, in the groups To verify: sparse points, To verify: high area outside every outline and To verify: improbable shape. Accept the real buildings, delete the rest, with the same buttons as the outlines. On a coloured cloud a green crown is not listed.

Roof 3D: pitches, ridges, dormers and chimneys

Build for this one (current outline) or Build for all (n) (every provisional and accepted outline, with a progress bar) reads the cloud points inside the outline and builds its roof: one plane per pitch, straightened on the axes of the building, with ridges and hips where two planes meet and the eaves on the outline. Each pitch is a closed 3D polyline, which you can edit, snap to and see in stereo. Dormers and chimneys are drawn as their outline at the elevation of their top. Building a roof again first removes the pitches inside that outline.

The status line gives the pitches, the dormers and chimneys and how far the planes are from the points, or why a roof was not built (too few points, no pitch recognised, pitches that do not explain the roof).

13.11.6 Correcting and accepting: the right-click menu

Select a provisional outline and right-click. The menu is the outline's own and works on the whole selection (right-clicking an already selected outline does not change the selection).

• Side handles. A selected provisional outline shows a handle at the middle of each side. Dragging it moves that side parallel to itself (perpendicular movement only), which is the quick way to put a wall where it belongs. Vertices keep the usual CAD grips, which snap to the cloud.
• Every vertex change from the menu, and from the vertex bubble, is one undo step.

13.11.7 Checking the outlines

Automation → Building outlines check… checks the outlines in the drawing — the accepted ones as well as the provisional ones, corrected by hand or not — and lists what is wrong. It also runs by itself after every generation and puts the result in the status line. The window lists, in this order:

1. Crosses itself
2. Overlaps another building (with the area in common; a house inside another's courtyard is not an overlap)
3. Not closed
4. Smaller than 3 m2
5. Under trees — to review (the orange outlines)
6. Provisional — to review (the other provisional ones)

Inside the two review groups the largest outlines come first. Click a row, or use < Previous and Next >, to centre the view on that outline and select it; Check again repeats the check after your corrections. Each row says whether the outline is accepted, provisional or under trees. Fix a crossing before you try to merge that outline.

13.11.8 Export to GIS: GeoPackage

File → Export buildings (GeoPackage)… asks where to save (it proposes export\edifici.gpkg in the project folder) and writes the outlines of the open drawing with their attributes. It needs an open drawing; with no outlines it says so. A wait window counts “Building i of n”. The file has two layers:

The polygons are 3D (they keep the elevations of the vertices) and a courtyard is a hole of its building, not a separate building; the area is net of the hole. The coordinate system is the project's (EPSG code, or the full definition when there is no code). A project with no declared coordinate system produces a file without one, and the log says so. Every feature has these attributes (the names are as they appear in the file):

The attribute names are in Italian in the file whatever the interface language. If the file cannot be created, or some buildings cannot be written, the status line says which and how many.

The command has no default key (assign one under Export buildings (GeoPackage) in Keyboard Shortcuts).

13.12 The command atlas (5.4)

The CAD commands are collected in a side bar by groups: Draw, Vertices, Trim and extend, Construction, Join and split, Cleanup, Transformations and Heights (Z). Pick a group, then the command. Each command has its own small dialog: what it does, the options it asks (distance, radius, angle, factor, tolerances), the step to do next (“Click the points; Enter or right-click ends”) and the buttons Start, Apply, Enter and Cancel (Esc). The tooltip of a command shows its command-line alias; right-click opens its dialog without starting it. A switch (Open with every command) decides whether the dialog opens each time.

The commands are the real CAD engine commands: the atlas only gathers them and lets you type the values in one place. Typed values are checked: a text that is not a number is refused and says so.

New in 5.4

• Vertices — Densify (a regular interval, or a fixed number of parts per side), Smooth (rounds a broken line by adding vertices), Simplify (removes vertices within a tolerance, keeping the shape), Remove duplicate vertices (coincident ones, or closer than the tolerance).
• Trim and extend — Lengthen (the end of a line by a distance, in its direction), Split at intersections (the selected lines are cut wherever they cross in plan), Clip with a fence (inside or outside a rectangle you make the fence) and Reshape (the part of a line between two crossings is replaced by a stretch you draw).
• Construction — Parallel at a typed distance, on the chosen side or through a point, and Stereo parallel, through a point picked in stereo with its height; they share one window.
• Move and Copy take the base point on the entity you click, with its elevation.
• Layers are listed in alphabetical order.

13.13 Heights (Z) of lines (5.4)

The Heights (Z) group changes the Z of the vertices of the selected lines. Select the lines first, then pick the command:

The command works on what is already selected, like every modify command. One undo step puts the lines back.

13.14 The CAD tutorial (5.4)

Open it from the Tutorial button of the atlas or from Help → CAD tutorial…. The real CAD commands run in front of you, one by one and in their variants (every form of circle, arc, ellipse, spline and polyline with arcs, then every editing chapter: vertices, trim, build, join, clean, transform, heights). A caption says what is being done and the result stays on screen before the next scene.

• Your drawing is not touched. The tutorial draws on a layer of its own, in an area apart, and when it ends or you stop it everything is put back exactly as it was.
• Controls: Space pause / resume, → next scene, ← see again, Esc stop, + / − speed. Closing the window during the tutorial first puts the drawing back.

The tutorial is also the product's check of the CAD commands: in the test build each scene writes the entities it produced, and a script compares them with the expected result in numbers.

14. Feature macros (rapid coding)

Feature macros bind a single key to “select a construction layer and start drawing a chosen entity type” — the fastest way to digitize themed features. Manage them in CAD → Macro Commands…: add new macros, rebind keys (press Esc to cancel a rebind), and Restore defaults. Macros are saved between sessions.

The built-in starter macros:

15. Spot Height Plan (piano quotato)

This tool seeds a regular grid of points over an area and reads each point's elevation from the cloud, producing a finished spot-height plan. Open it from Tools → Spot Height Plan…, the toolbar Quote button, Ctrl+Shift+H, or the command SH.

15.1 Define the boundary

Click Draw boundary and pick vertices in the viewport; right-click to close the boundary (at least 3 vertices). Use Undo last or Clear as needed, then Done. You can Redraw at any time.

15.2 Grid & seed

• Link X/Y step — keep the X and Y spacing equal (default), or set Step X and Step Y independently.
• Seed — the grid starts from the boundary centroid by default; use Pick seed to set a custom origin point (and Reset seed to revert).

15.3 Elevation from the cloud

• Z method — Nearest point, Lowest (ground), or IDW average.
• Interpolate holes — where there is no cloud beneath a grid node, estimate its height from neighbours instead of leaving a gap.
• Remove spikes — replace anomalous heights (e.g. a power cable over a road) with an interpolated value; set the spike threshold in metres.

15.4 Symbol & layer

• Marker — Dot, Cross, Circle or Square.
• Auto size — when on, marker size and quote text height are set to 0.1 × step; otherwise set them explicitly.
• Decimals — precision of the elevation label (0–6).
• Use custom block — draw your own block at each point instead of the built-in marker.
• Layer and ACI color — the construction layer (default Spot Heights, color 7).

15.5 Seed, verify in stereo, confirm

1. Seed points — a progress dialog samples the grid from the cloud and can be cancelled.
2. Preview — seeded points are shown temporarily, color-coded: cloud-derived, interpolated, spike-corrected. Click a point to delete or restore it. This is where stereo verification pays off — suspicious points are easy to spot in 3D.
3. Confirm & draw — the verified points are drawn as final geometry on the construction layer. Or Re-seed / Cancel.

All Spot Height settings persist between sessions.

15.7 Polygon Cleanup

Cloud → Polygon Cleanup… (Ctrl+Shift+J) hides cloud points inside a closed boundary drawn in the current view: no plane fit is required, the tool uses the oriented view frustum and a polygon you trace on screen.

1. Open the panel from Cloud or press Ctrl+Shift+J.
2. Draw boundary — left-click vertices on the cloud in the main viewport; the polyline closes when you have at least three points.
3. Preview — points inside the extruded volume are highlighted before anything is hidden.
4. Apply hide — matching points move to the temporary hidden cleanup layer. Clear hidden restores them.

Use this for quick clutter removal in the current oblique or zenith view when a planar wall fit is not needed.

15.7.1 Hiding what is in the way, with Q (4.7)

While you are drawing, a tree or a building standing between you and your subject has to go away in one gesture, not through a panel. Hold Q and drag a rectangle in the main view:

• Left to right — the band is green, and that area is hidden.
• Right to left — the band is red, and it gives back the areas it touches.

It is the same direction convention as CAD selection, so there is nothing new to remember. Rectangles add up: draw several and the hidden region takes the shape you need instead of a convenient box. Release Q and you are back to drawing; press it again and you carry on from where the masks were.

Shift+Q clears them all, without opening anything. The broom button beside the orientation triad turns the whole cleanup off and on, so you can compare what you hid with what is really there and go back.

The outlines of applied masks are drawn only while the Polygon Cleanup panel is open. During work they were in the way and told you nothing you could not see.

The masks belong to the project that defined them. Opening another project and closing it no longer leaves its masks in force over this one — a case where the panel said there was no cleanup while the cloud was full of holes.

15.8 Area of interest

The area of interest says where the commands compute on the cloud. On a large project it is the way to work one block at a time: a command that reads the whole cloud in minutes reads the area in seconds. The cloud is always drawn whole; the area only bounds the work.

Open it with the green Area of interest button in the column at the top left of the mono view, from Cloud → Cleanup & area of interest → Area of interest, or with Ctrl+Shift+A. The window says which area is set (a polygon with its vertices and square metres, a rectangle, or none) and has two buttons:

The polygon is drawn green on the cloud and is saved with the project (in analysis/cloud_aoi_polygon.json, next to the rectangle around it in the project file).

Which commands use it. Every command that computes on the cloud: with an area it works inside the polygon, without one on the whole cloud — never on nothing. Next to the button that starts a command a line says which: Area: the area of interest (n m²) in green, or Area: the whole cloud.

Commands that take an area of their own (volumes, spot heights, sections, polygon cleanup, change class, click tools) keep it.

A rectangle can still be drawn from the Scene Tree (right-click on a cloud: New AOI + draw rectangle); drawing it replaces the polygon.

15.9 Elevations & Sections — plans, elevations and sections from the cloud

A point cloud already contains every drawing you are going to deliver; what it lacks is a frame to cut them out of. Tools → Elevations & Sections… (Ctrl+Shift+B) builds that frame: a work box aligned to the object rather than to north, whose faces and cut planes become plans, elevations (prospetti) and sections. Every one of them is rasterised from the points themselves, so an orthophoto exists even where nobody ever took a photograph.

The session is filed in the Scene Tree under the cloud it was built from, and it stays there: you can close the panel and the box, the products and their rasters survive with the project.

15.9.1 The work box

1. Opening the panel never touches the mono viewport on its own — a click there does whatever it already did (orbit, pick, nothing) until you press Draw AOI.
2. Draw the area of interest on the cloud. Left-click marks vertices, right-click or Enter closes it (three vertices minimum), Backspace undoes the last one, Esc clears.
3. The box is then fitted from the cloud points inside the polygon — not from the polygon edges. Principal planes are detected on the points, so a building comes out with its own walls square to the box even when it stands at an angle to the survey grid. The panel reports the result as X' × Y' × Z in metres.
4. Redraw AOI polygon… recomputes it if the fit caught the wrong thing (a neighbouring roof, a hedge) — this one re-arms drawing immediately, no need to press Draw AOI again. The panel says what went wrong — no points in the AOI, or no detectable principal orientation — rather than fitting something arbitrary, and drawing is NOT re-armed automatically after a failed fit: press Draw AOI again when ready.

15.9.2 Products

Products live in folders in the Scene Tree, and each one is a view taken off the box:

15.9.3 Where the image is made

The raster is not made here. Choosing the face, the kind of image and the resolution, and pressing Generate, all live in the Workbox panel, which is the one place a drawing is produced; the tone adjustment lives in the viewer, next to the image it changes; and putting a drawing on a sheet is composition, done from the sheet among the images already made. This window defines the box and its cuts, and stops there.

15.10 CAD Sheets — the final drawing on paper

A CAD Sheet is the delivered sheet: a real paper size, a real scale, the drawings placed on it and your CAD drawn on top. It is a drawing of its own, separate from the project CAD, saved with the project and filed under Products — which is why it has its own export rather than riding on File → Export.

Two jobs, two places, and they are deliberately not the same window: you generate a drawing in the Workbox, where the face, the kind of image and the resolution are decided; you compose it on a sheet, choosing among the ones already made.

The sheets are the Delivery step of every workspace — aerial, terrestrial and mobile — so they are on the CAD menu everywhere: CAD → CAD sheets makes a new sheet, lists the sheets of the project (the open one is ticked; pick another to open it) and closes the open one. The Scene Tree and the navigator below reach the same sheets.

The Sheets step

In the activities that make drawings from the cloud — Bridge, Tunnel, Building, Frontage and Interiors — the sheets are also the last step of the activity bar, Sheets, after the plans, elevations and sections it composes. Clicking it opens the CAD sheets navigator. The step is done when a sheet of the project carries at least one drawing (a closed sheet counts too); it waits for the drawings of the work before it lights up.

15.10.1 CAD sheets navigator

View → CAD sheets docks the navigator in the pane beside the mono view — on the Terrestrial Mapping workspace, the pane an aerial job gives to the Zenith Overview. A facade has no planimetric map worth looking at; it has sheets to compose on, so on that workspace the Overview button is not offered and this one takes its place. The Overview itself is still one menu entry away.

• The list — every sheet in the project. The one that is open is amber; the one you have chosen carries the selection bar. They are two different things.
• + beside the title makes a new sheet. The context menu on a row offers Open, Close, New CAD sheet… and Delete — a sheet with drawings on it asks first, saying how many.
• The figure — the chosen sheet drawn small: the paper in its true proportions and one rectangle per drawing placed on it, at its real position and size. It is a layout, not a preview of the CAD: to paint the CAD a sheet has to be opened, and the engine holds one drawing at a time.
• Open — opens the chosen sheet for editing. Clicking a row only chooses it: opening spreads the sheet window across the screen, which is the opposite of a glance. A double click opens, as lists do everywhere.

The figure is available for closed sheets too, read from the sidecar beside the .vec. A sheet that has never been saved has nothing to draw yet.

15.10.2 Drawing in metres on the orthophotos

The usual job on a sheet is to place one or more orthophotos from the Workbox — a facade, a plan, a section — and draw on them. A new sheet is therefore drawn in real metres: one drawing unit is one metre in the plane of the orthophoto, an orthophoto lands at its true size on the ground, and every length, coordinate and dimension you read on the sheet is in metres. The scale 1:N belongs to the paper, not to the drawing: it decides how big the paper frame is around your drawing, and how the sheet prints and exports.

• The paper is drawn by the sheet window — white, with its shadow and black edge — at its size in metres: an A3 at 1:100 is a 42 × 29.7 m frame. It is a guide for the layout and is not part of the drawing.
• Text and dimensions are created 2.5 mm high on the paper: 0.25 m at 1:100, 0.125 m at 1:50. They follow the scale every time the sheet opens or its scale changes; text and dimensions already drawn keep their size.
• The status line of the sheet says drawing in metres.

The sheet window

A sheet is born with a default — A0, 1:100, 600 DPI — and opens with its settings closed: you open it to draw on it, not to configure it. Along the top: Save, Close and one command, Add a drawing…. Middle button pans the sheet.

Drawing and editing on a sheet

While a sheet is open it is the drawing the CAD commands work on, exactly as the project CAD is the rest of the time:

• the command line aliases — L, PL, C, A, M, CO, E, TR, OFF, X and the rest — and the Draw / Modify menus and toolbars;
• Space repeats the last command, Ctrl+Z / Ctrl+Y undo and redo;
• Object snap and polar tracking act on the sheet, with the choices you made for the 3D drawing. Cloud Z does not apply on paper: turning it on while a sheet is open is refused and the status line says why, and your Cloud Z choice for the 3D drawing is kept.

Adding a drawing

Add a drawing… lists the images that exist, one row each, with the drawing's name, the kind of image and its ground sampling:

Y' min    RGB        4.2 mm/px
Y' max    Depth      8.5 mm/px

Choose one and click the sheet to place it. One at a time. The same list is on the right-click menu over the sheet, and it is the same list — not a second road with its own behaviour. While you are looking at a drawing in its viewer, it goes on a sheet from there: Deliver… → On a CAD sheet.

Nothing is regenerated: the size on the sheet comes from the drawing's ground extent (its true size in metres, or divided by the print scale on a sheet in paper millimetres), never from the raster's pixel count, so a coarse image and a fine one land at the same size — one is simply less dense. When an image is coarser than half of what the scale would print, the row says so in millimetres before you pick it. For a sharper one, regenerate it in the Workbox at that resolution.

Placing other CAD drawings

At the top of the same list, CAD drawings:

• Project drawing (1:N) — the project CAD as last saved, at its true size;
• Sheet name — another sheet of the project as last saved, at its true size on the ground whatever the scale of either sheet (its vector drawing only: its orthophotos stay on that sheet);
• DXF/VEC file in metres (1:N)… — a survey drawing;
• DXF/VEC file already in millimetres (1:1)… — a title block or a legend drawn in paper millimetres, placed at its size on the paper.

The drawing arrives as one block, centred on what the sheet window is showing, and selected: M moves it to its place, X (Explode) turns it into ordinary lines, polylines and texts you can edit. A DXF is first converted as the CAD import does, with its progress window.

Moving, cropping, removing

Right-click a placed drawing for Move (drag the orange centre handle), Crop (drag the blue edge grips — cropping never rescales), Reset crop, Open in image viewer, Adjust and Delete. Esc leaves the mode. Outside those modes the cursor is the normal CAD cursor: draw, dimension and annotate on the sheet exactly as in the model.

Changing the sheet, and what follows

Sheet settings holds paper size, orientation, Scale 1:N and Export DPI. Apply commits them.

15.10.3 Export

Export sheet… writes the sheet as DXF, or DWG where the DWG converter is installed, and files the result under Products with the date it left the application. The Scene Tree marks a product delivered or on a sheet once it has, so you can tell at a glance which drawings have already gone out.

A sheet with orthophotos goes out as DXF. The DXF of a sheet is written as DXF 2000, in the sheet's units: metres for a sheet drawn in metres (plot it at 1:N), paper millimetres for an older sheet. The DWG declares the same units. When the sheet carries orthophotos, each one is written at full resolution (not the reduced copy shown on screen), with the tone adjustments you see on the sheet, as a PNG with transparency for empty cells, in a folder next to the drawing named <name>_ortho. The DXF refers to the images by file name, on the layer ORTOFOTO, and each image is clipped to its outline on the sheet. Keep the DXF and its _ortho folder together, as you would an AutoCAD or ZWCAD package; to get a DWG, open the DXF in AutoCAD or ZWCAD and save it as DWG. When the export finishes the status line says how many orthophotos went into the folder.

15.10.4 Workbox panel

Changed in 5.2. Kind of image is in the order RGB, Intensity, Depth, and a new drawing starts as RGB. On a cloud without colour (or intensity) the drawing moves to the first kind the cloud can give. Drawings already saved keep their kind.

View → Workbox docks the drawing set of the active work box on the Terrestrial Mapping workspace — the same pane the CAD sheets navigator and the Multiview use: one pane, one occupant at a time. Every occupant has its own button over the mono view, and a panel that is open always keeps its button even where the job does not otherwise offer it, so nothing can be left open with no way to close it.

Always in view: the volumes

• Volumes — every work box in the project, one row each, with its size and, in a second muted line, whether it isolates the cloud and how many zones have been hidden inside it. The current one is highlighted: what you hide, clean or draw belongs to it. No volume is a row like the others and means the whole cloud. + beside the title adds one; the context menu on a row deletes it, asking first when it carries drawings.
• Only the volume — hides everything outside the box. It follows the operator, not the box: switching volume keeps it as you left it.
• Hide clutter volumes… — the Polygon Cleanup tool, right where the volume is. Draw an outline in the view and the points inside it disappear. With a volume current the hidden zone belongs to that volume; with none, it belongs to the cloud.

Four tabs follow: Drawings, Images, Volume, Resolution.

Drawings

The six faces of the box are six buttons, in two rows of three: front, back, left, right, top, bottom. The one you are working on is lit. Pressing one selects that drawing and takes the view to it — the same move as the preset views on the axis triad, so there is no second way to orient the model.

The fit picks the box's principal axis but cannot know which of the two long sides is the facade, so Front on the other side sits next to the buttons: it turns the names through 180° — front becomes back, left becomes right — and the views follow. It belongs to the volume and is saved with it.

Sections and intermediate plans are not faces: they appear in a dropdown under the buttons, and only when they exist. + Section offers the three cuts by what they do — cross, longitudinal, horizontal — and + Plan adds a plan. There is deliberately no add elevation: the four elevations are the faces of the box. To delete one, right-click its row in that dropdown; the elevations do not appear there, so the command never has to refuse.

What the drawing takes

All of it. The slice is the whole volume, and what ends up in front is decided by the rasterizer, which orders the hits by distance from the viewer and keeps the nearest — painting back to front. There is no depth window to set, and no facade to measure first.

What you do not want in the drawing is taken out afterwards, with Hide clutter volumes…: a decision about what to throw away, not a geometric law applied blind. Hidden zones apply to the drawing too — the same rule the renderer uses on screen — so a point hidden in one view cannot reappear in the other.

Which way it looks is read from the camera at the moment you press Generate: you draw the face you have in front of you. Elevations and plans only — a section has no front and no back, and looking at it from the other side does not turn it into a different section.

Working on the selected drawing

• Name — type a real name (A-A, North elevation, Plan +3.20). It is saved when you leave the field, and the Scene Tree leaf follows.
• Kind of image — Depth, a ramp by distance from the plane, which every cloud can do; RGB, the cloud's own colour, which is the photo-plan you trace over; Intensity, the laser return, which reads the material and does not depend on the light of the day you scanned. Each kind is a separate image with its own file: generating one does not replace the others, and switching kind points the drawing at that kind's image when it already exists.
• Close the windows — see below. Off unless you ask.
• Generate — produces the image and opens it. It is the only command of this tab: opening an image is done in Images, and deleting a drawing on its row.

Generate writes the image raw, without the tone adjustment. The file on the bench is the measurement; brightness and contrast are a lens, set in the viewer while looking at the image, and they are baked in when the image leaves — at delivery. Baking them here as well would mean the viewer applied them a second time on top of the first.

Close the windows

A laser does not come back from glass, so every window is a hole in the drawing — a big one: at 4 mm/px a 1.2 m window is 300 pixels across. This is not a defect of the cloud or of the rasterizer; those points do not exist.

Close the windows fills them from their own border. It is off by default and asked for each drawing, because it invents pixels, and three rules bound what it invents:

• the hole must be closed — no pixel of it touching the edge of the drawing. Without that rule, it would fill the sky;
• it must be smaller than 4 m². A carriage entrance or a courtyard seen through an opening is a closed hole too, and those are not glass;
• the value comes from the border of that hole — the first solid pixel left, right, above and below, weighted by distance. The depth is therefore the wall around it, which is honest because the wall is there; the colour is an average of the border, and that part is invented.

The filled pixels stay marked as interpolated: nothing counts them as measurement.

Images

The images that exist on disk, one row each: drawing, kind, pixels and ground sampling, with * marking one already delivered. Click a row to open it; right-click for Open and Delete the image. Deleting asks first — a raster costs minutes — and takes away only the file on the bench: a copy already delivered under exports is in somebody's hands and this command does not take it back.

Volume

What is decided once per box: its size and Edit…, which reopens the Work box window to redraw the AOI, and a line saying where what you hide now ends up.

Resolution

Nothing is generated here. You declare the resolution deliveries will use, and read what would come out of it. It is the only place that resolution is declared.

• By scale — 1:20 to 1:1000, with the export DPI. The line below translates it: one pixel is so many millimetres on the ground.
• By pixel size — a bare ground sample distance instead. A GeoTIFF for a GIS has no print scale, and forcing one on it would invent a dependency that does not exist.
• What will come out — one row per drawing: its size in metres, the pixels it will produce at the chosen resolution, and what it weighs in memory while it is being made, with the total. Anything past the 20,000 px limit is marked in red with what to do about it — a smaller scale, or splitting the drawing.

These settings belong to the volume and are saved with it.

15.10.5 The viewer: adjusting, and delivering

Generate opens the image in the Image Viewer, because that is where the next decision is taken: you look, you adjust, you hand over.

Adjust

Adjust in the toolbar opens a strip above the image — not a separate window, so you watch the result while you move the number. Its name turns amber while an adjustment is active, so you can tell from outside that the image is not raw.

• Brightness and Contrast — drag, or Ctrl+click to type.
• Auto — takes the 2% and 98% of the image's own tones to black and white, counting only the content: the rasterizer's empty background does not vote, or on a facade with a lot of empty around it the background would decide. When the image has too little content, or all of it on one tone, it says so instead of inventing an adjustment.
• Local — contrast tile by tile instead of over the whole image. This is what brings out a facade that is in the sun on one side and in shadow on the other.
• Reset — back to the image as it was made.

The numbers belong to the drawing, not to the viewer: they are saved with the project and come back when you reopen it. The line under the strip says which of the two cases you are in: these values go into the delivery, or screen only.

The adjustment is recomputed from the pixels as loaded, never from the previous result, so taking the contrast to 2 and back to 1 returns exactly the image you started from. On an image larger than 4096 px the viewer works on a reduced copy: at full resolution each move of the slider cost half a second and once produced a stall of nearly seven. Nothing that matters is lost — the screen cannot show 67 million pixels, and the numbers travel to the delivery, which regenerates from the cloud at full resolution.

Deliver

Deliver… sits in the viewer's menu bar, next to the image you are judging. It makes a file: GeoTIFF, PNG or JPEG. It states the resolution declared in Resolution and how many pixels that gives, and it is not where you change it.

The file on the bench is raw, so a delivery carrying an adjustment goes through a regeneration from the cloud with those values baked in — the same road a delivery already takes whenever the resolution does not match the file on disk. Delivered files go to exports/<volume name>.

From the viewer to a sheet

Under the files, Deliver… lists On a CAD sheet: the sheets of the project (the open one is ticked) and New sheet. Choose one: the sheet opens — a new one is drawn in metres — and the image you are looking at is ready to be placed: click the sheet where it goes. It lands at its true size; nothing is regenerated and no scale is taken from the sheet. The entry is greyed, saying so, until the image has been generated.

15.11 Railway

For a railway survey the program finds the two rails of each track from the points alone, measures them every 2 m, and then finds the parts of the overhead contact line: masts, cantilevers, messenger wire, droppers, feeder and anchor spans. Every step follows the same rule as the rest of the product: the result is first shown as a draft on the cloud, you check it, and only Confirm writes it into the drawing and into a table. Nothing is interpolated: where the cloud has no points, the stretch is declared as not measured.

The commands are in Cloud → Railway: Find the track…, Calculate the track table, Track table settings… and Track sections and profile…. The other steps — masts, cantilevers, messenger wire, contact wire, droppers, feeder, and the objects along the line (tunnels, cabinets, sleepers, 15.11.5) — are steps of the railway chain (8.7.5): Run the chain starts them in order, first the steps that find objects, then those that only write deliverables. The Track view of the bottom panel shows the step you chose. None of these commands has a default shortcut; each can be assigned one in Keyboard Shortcuts.

All the clouds that are on are read together, as one cloud: a track that runs from one cloud into the next is followed across the border, and masts and wires are looked for in every cloud.

Layers written. Their names follow the interface language; below they are given in English, with the Italian name. A project made before keeps its layers under the Italian names: the program reads them and replaces them, it never writes a second copy beside them.

Each confirmation empties its own layer and writes it again, so repeating a step replaces the previous result instead of piling a second copy on top of it.

15.11.1 Find the track

Open Cloud → Railway → Find the track…. The command starts by itself with a search over the whole corridor: it reads the cloud along the line (about a second), finds every pair of rails that continues along it, follows each one and measures its rail heads (about 5 seconds per track). A progress bar shows Following track i of n and the metres followed; Esc cancels. The tracks already on the layer are left out, so the same command can be repeated to find the others.

The program uses the gauge and the rail head width declared for the project; when the project declares none it uses the standard values (1435 mm and 72 mm) and says so in the dialog.

What you see

The result is a list, one row per track, in green or amber:

• Green — a track you can confirm: its length in metres and the number of stations whose rail heads were measured out of those followed. To be sure the rail heads must be measured on at least 90% of the followed stations, and the length must be reproduced when the track is followed again from the other end (the two ends agree within 10 m, 98% of the length covered).
• Amber — a track that is not sure, with the reason. It is not written as a track. Typical reasons: the cloud is too sparse to measure the rail heads, the track was lost after a number of metres, the length is not reproducible. A track that is not sure because of a gap is cut at the gap, and each piece is judged on its own (T0a, T0b…): a piece can be green while the other is not. The gap stays a stretch not measured.

Click a row to move the view to that track. Then press Confirm the sure tracks (n) to write them, or Discard. The lines go on the layer of the rails (RAILS), two polylines per track, one per rail.

Rails seen alone

What is seen is documented, even when the track is not complete. A rail whose other rail was not seen (hidden, or out of the survey) is listed as Rail n seen alone: m metres seen; the measured stretches of an amber track are listed the same way. Write the rails seen (n) writes each as a polyline through the rail heads measured, broken where it was not seen, on RAILS_SEEN_ALONE. No rail is added and no rail is guessed from the gauge: a rail seen alone is not a track, and the table and the sections need both rails. Where the contact wire of its track is seen, the wire tells on which side the track lies, and the axis of that track, half the gauge beside the rail, goes on LINE_AXIS in orange.

After the confirmation: the line axis and the platforms

When tracks are confirmed, the program follows the line beyond each track and looks for the platforms along it, under the progress cover (i/n; Esc cancels, and the tracks stay written). Every 2 m the axis is placed by what is seen there, and coloured by it on LINE_AXIS:

In a tunnel the axis is placed between the two walls, at the offset learned on the sure track. The axis never comes nearer than 2 m to the axis of another track. It is broken where the source changes and where two points are too far apart. The offsets of wire, platforms and walls are learned on the sure track, not standard values written in the program.

The platforms go on PLATFORMS, in yellow, at the height of their top: the edge facing the track (a vertical step with a level top, higher on the outside), 0.8 to 3.5 m from the axis, at least 10 m long; a low pavement from 10 cm above the rails counts. The status line gives the metres followed beyond the tracks and the platforms found.

The masts and the wires (15.11.3) are then looked for along the whole line axis as well, so that supports are found in stations and beyond the sure tracks.

The stretches not measured

Under a track the dialog lists the stretches that could not be measured (up to eight, then …and n more): a station that did not validate, with its reason, or a gap that the search did not cross (gap of n m, not crossed). A click on a row takes the view there. They are never interpolated: in the table they stay empty with the reason.

Find the other tracks

On a double or multiple track, once one track is on the layer the button reads Find the other tracks (otherwise Find all the tracks). It repeats the corridor search leaving out what is already written. If everything is already on the layer it says so.

If the search finds nothing: click on the track

If the data is not a long corridor (the message gives the ratio of the axes), if no pair of rails continues along it, or if you cancel, the dialog goes back to the click: click on the track in the plan view, even just near it. The program proposes the two rails (Rails found, 1.5 m between the heads, direction n degrees) and, if another pair of rails lies near the click, tells you at what distance. Enter (or Follow the track) follows the track; another click moves the proposal; Esc discards it. The outcome is green (confirmable), amber (not sure: you can only discard it, click again or use the polygon fallback) or red (no track, with the reason in words: no points near the click, nothing at rail height, no pair of rails at the gauge, and so on).

Where there is a second level under the track (a bridge or an underpass) the rails cannot be recognised from the click if it falls on the lower level: click the track outside the bridge, and it will be followed across it.

The polygon fallback

When the outcome is amber or red, Draw the rails with polygons turns the view to the section at the place the search found. In the tab Search by section (15.12) press Rail, draw a polygon around the head of each rail (left click for each vertex, right click to close), then Start search. The lines go on ROTAIE as before and the track table can then be calculated from them.

15.11.2 The track table

Cloud → Railway → Calculate the track table takes the rail lines on the layer (ROTAIE, or the layer chosen in the settings), pairs them two by two, reads the cloud again around each rail and fits the head of each rail every 2 m. The same step is offered at the end of Find the track (Compute the table) and as a button in the Track view.

Cloud → Railway → Track table settings… opens the dialog Rail geometry with the Layer of the rails, the Head width (mm) (for a 60E1 rail it is 72) and the result of the last table. The head width is declared by you and never read from the data; with 0 the gauge is not written.

Files, in exports/rail_geometry/: rail_geometry.csv (the table above) and, when two or more tracks run side by side, rail_track_spacing.csv: the distance between neighbouring tracks, axis to axis in plan, with the height difference, for tracks within 10 m of each other and never across a gap of more than 6 m between valid stations. In an interface language other than English the files use the semicolon and the decimal comma, so that Excel opens them directly.

The Track view

When a table is calculated the bottom panel opens the Track view (it is shown in the toolbar while it is in front; Back to the deliverables returns to the Deliverables view). At the top the buttons Find the track, Calculate the table and Settings… run the same commands as the menu. Below:

• the summary: tracks, stations, measured, not valid; the median and noise of the distance between the heads and of the cant; the derived gauge; Jumps over 12 mm between neighbouring stations (green when none); the distance between the tracks; the folder with Open folder;
• two charts along the track, Distance between the rail heads (mm) and Cant (mm); hover to read the value of the nearest station, or the reason it is not valid;
• To look at: the stations that did not validate and those more than 12 mm from the median, up to six, each with a Go button that frames the station in the view. If there is none, the view says so in green.

15.11.3 Overhead contact line

The parts of the overhead line are found one after the other from the tracks, in the Track view, each in its own collapsible part. Every step has the same three moves: Find (the draft is drawn in the view and listed, a row takes the view there), Confirm (writes the layer and the tables), Discard. Each needs the previous one: the program tells you if a mast or a track is missing. The tables go to exports/<folder>/ and are read by the Deliverables view as the Mast inventory, Cantilever inventory, Messenger wire geometry, Contact wire geometry, Dropper inventory, Feeder geometry and Anchor inventory.

The contact wire, seen or not. Whether the cloud holds the contact wire depends on the survey. On a mobile survey from track level close under the line it is usually a clear, continuous line, and Find the contact wire measures it: on a test survey of an Italian line it came out at 4.9–5.0 m above the rail, almost level from one support to the next. On other surveys the scanner returns neither the contact wire nor the lower part of the droppers: Find the contact wire then reports that no contact wire is seen and draws nothing. In both cases Find the droppers also gives, for every span with droppers, the lowest visible end of the droppers as an upper limit: the contact wire is at that height above the rails or lower. A span where the messenger wire is measured but the cloud holds no dropper is listed as such (not in the cloud), it does not disappear.

15.11.4 Track sections and profile

Cloud → Railway → Track sections and profile… opens the dialog Track sections and profile. The sections are cut in the cloud itself (no terrain model is needed): a slice 0.5 m thick along the axis, one height every 10 cm across it (the median of the points of each 10 cm cell); a cell with fewer than 3 points is a gap and is left open. It needs at least one track on the layer of the rails.

The choices are saved with the project. Press Run: a progress bar shows Cutting the sections (i/n). The result is n sections on SEZIONI, profile of m track(s) on PROFILO_FERRO; sections that found no point of the cloud are left empty and counted. SEZIONI holds the section lines in 3D; PROFILO_FERRO holds the profile, the axis of each track at rail level. Both layers are emptied and written again at every run.

Tables, in exports/rail_sections/: sezioni.csv (section number, axis, chainage s_m, lat_left_m — positive to the left of the axis, looking along the track — east, north, z) and profilo_ferro.csv (track, s_m, east, north, z and slope_permille between consecutive stations).

15.11.5 Tunnels, cabinets and sleepers

Three steps of the railway chain look for objects along the line axis (LINE_AXIS; without it the step says first Find the track). Each runs under the progress cover (Esc cancels, nothing is kept), on all the clouds that are on. The result is a draft drawn in the view, with a summary of what was measured; tunnels and cabinets are also listed one per row, with the same number in the view, and a click on a row takes the view there. Confirm empties the layer and writes the draft; Discard drops it.

15.12 Search by section

For a linear element — a wire, a rail, a kerb, a platform edge, a fence — you draw its section once, on a vertical plane across it, and the program follows it along the cloud in both directions, station after station, and draws the lines it finds. It is the same engine that finds the rails and the wires by themselves, but here you decide the element.

Where it is

It is the tab Search by section of the Reconstruction view of the bottom panel. It opens from the row of the Deliverables view whose method is Search by section (the lines then go on the layer of that row), or with the shortcut Search by section (unassigned: assign one in Keyboard Shortcuts). It needs a project with a cloud that has an octree and an open drawing. On a railway project the tab also offers Find the track and the Rail button (see 15.11.1); on other types of work they do not appear.

The steps

1. New search, then 1. Direction. Press Draw direction and click two points on the view along the element: a temporary elastic line follows the cursor and nothing is written into the drawing; Esc takes back the point. Or select one polyline that follows the element and press Use selection; the polyline is not changed. The section is placed halfway along the line.
2. 2. Section. The view turns to look along the element and shows a 2 m slice of the cloud. Snap and stereo work as usual. Draw on the layer SECTION_SEARCH (the name is in the interface language):
   • Point — a point on each wire or conductor; the search follows the densest returns around it.
   • Polyline — a polyline across a rail head or a kerb; each corner is followed by the shape of the points around it, not by their density.
   • Rail (railway projects) — a closed polygon around the head of a rail: left click for each vertex (at least 3), right click to close it and go on with the next rail, right click with no vertex to finish; Esc takes back the last vertex, Undo last removes the last rail. The head is read from the points inside and the search starts from it. A polygon that does not hold a rail head is refused with the reason (too few points, it crosses itself, what is inside is not a rail head and is some number of millimetres wide, the polygon cuts the head: draw it a little wider).
   • Other profile — the same gesture for a kerb, a platform edge or a fence: the search starts from the centre of the points inside. Nothing is checked about what they are: the polygon is your word.
   The line Drawn: n point(s), m polyline corner(s) and Rails and profiles taken: n say what the search will start from. Elements that lie off the 2 m slice are counted: only their position across and their height are used.
3. Start search. The search runs on every core while the view keeps drawing, and the lines grow on the plan in yellow as they are found. Forward n m, backward n m (n stations) shows the progress; the bar only says it is alive, it does not pretend a percentage. Cancel stops it and nothing is written.
4. The result. One 3D polyline for each drawn vertex is written into the drawing: profiles (rails, kerbs, edges) on the layer of the row or on SECTION_LINES, wires on the layer of the overhead lines. The summary says how many metres were followed, how many lines were written, and why each side stopped:

To go on beyond a stop, press New section and draw a new one past the point where it stopped. A gap in the cloud is crossed only when the engine can bridge it with the shape it has been following; a stretch that cannot be bridged ends the line, it is never filled with an invented one.

15.13 Search from a sample

You show one object that repeats — a sleeper, a mast, a pole, a manhole, a signal — and the program finds all the others along a route or in an area. It is the tab Search from a sample of the Reconstruction view, opened from the row of the Deliverables view whose method is Search from a sample (for example sleepers, masts, electrical cabins and signals on a railway); the instances are written on the layer of that row, otherwise on SAMPLE_SEARCH.

1. 1. Where to search. Select in the drawing one of:
   • Selected polyline = route: the objects are searched along it, within the Corridor half-width you type (for example 10 m for masts beside a track, 0.5 m for the sleepers of one track); the sample box follows the direction of the route. The route length is shown.
   • Selected closed polyline = area: the objects are searched inside it.
   • Area of interest: the area of interest set on the cloud.
2. 2. The sample. Press Draw sample and click two opposite corners of ONE object in plan, a rectangle a little larger than it (an elastic line is shown; nothing is written into the drawing; Esc takes back the first corner). Or select a closed polyline around it and press Selected closed polyline = sample. The sample must be at least 10 cm per side and hold enough points. The program shows its size and Bottom and Top heights taken from the points: a wire passing over the object is left out, and you can type other heights.
3. Minimum likeness (default 0.50): 0 is an ordinary place, 1 is the sample itself. 0.50 finds what is closer to the sample than to an ordinary place; lower it to find more (and to delete a few false ones).
4. Search. It reads the cloud of the route or area on every core and compares every place with the sample, with a progress that names the phase (Reading the cloud of the route, Looking at ordinary places, Comparing every place with the sample, Placing each instance); Cancel stops it.

The result

The summary says n found (m deduced from the spacing, k off the spacing) and, when the objects are regularly spaced (sleepers, poles), the regular spacing in metres. The instances are drawn on the view and listed in three groups, each place with its likeness and bottom height:

• Found — recognised from the cloud.
• Deduced from the spacing — a place where the regular spacing says an object should be although the cloud does not show it clearly; shown in orange.
• Off the spacing (check) — found, but not where the rhythm of the others would put it; shown in red.

Every row has a Keep box; Keep all and Keep none set them together. Write n into the drawing writes one closed rectangle per object kept — its footprint at the bottom of the box — with a progress bar; Discard writes nothing and keeps the sample and the route for another search. Change where goes back to step 1. If nothing resembles the sample more than an ordinary place, the program says so: lower the minimum likeness or take a more typical sample. If the corridor or the area is too large it asks you to narrow or split it.

15.14 Object reports

When you see on the cloud a particular object to document — damaged barrier, unknown object, a tree too close to the line — you can make a report on it: a record with its position, a picture of the cloud from above and from the side, a category and a note, optionally an identification proposed by the AI, and, only when you confirm, a mark in the drawing. It works on any data. It is the tab Objects to identify of the Reconstruction view, also reachable from the row Report objects of the Deliverables view, where it is optional: the chain is complete without it.

1. Report object. Optionally type a category first. Press Report object and click the object on the cloud (right click cancels). The program creates the report in the folder segnalazioni/S0001/ of the project (S0002, … for the next ones), with the world position, the time, and the pictures: the cloud from above and from the side, with a red ring on the reported point, and, if the project has a mobile run with panoramas, the crop of the nearest panorama (a station between 2 and 40 m from the object). If a picture cannot be made, the report is still created and the missing pictures are named.
2. Describe it. Select the report in the list (arrow keys go through them): you can edit the Category, the Label (what the object is) and the Note. Go to frames the object in the view.
3. Identify with AI (optional). It sends the pictures of the report, its position and your note to an AI vision service through our licence server, and it uses AI credits; Identify with AI (all to identify) does it for every report still waiting, in the background, and you can stop it. The first time you must accept that the pictures are sent (I agree, asked once on this computer); nothing is sent until you press an AI button. The answer is only a proposal: it appears marked (AI) with its confidence, and Accept takes it as the label, which you can still edit. Without consent, licence session, credits or network, everything else keeps working and the disabled button says why.
4. Confirm. It needs a label, a category or the AI proposal. Only now is a marker (a cross with the report number and label) written into the drawing, on the layer OBJECT_REPORTS (the name is in the interface language). Reopen removes the marker and makes the report editable again; Delete removes the report with its pictures and asks once more.
5. The list is grouped by label, with the status of each report: To identify, Identified, Confirmed; the header counts reports, those to identify and those confirmed.
6. Export reports writes in the segnalazioni folder of the project segnalazioni.html (a page with the pictures, readable and printable) and segnalazioni.csv with its segnalazioni.xlsx copy.

The reports are files of the project: they are read again when the project is opened.

15.15 Structures: tunnel, underpass, bridge

One job type, Engineering structures (tunnel, underpass, bridge), covers three kinds of work. Tick it under What do you want to produce? in the New Project dialog; types of work can also be added later from the Deliverable view. Inside it you choose which structure the project is about: Tunnel or underpass, or Bridge (the Structure choice at the top of the panel). The job chain, the panel and the list of drawings then show only the entries of that structure.

A tunnel and an underpass are the same work: same panel, same drawings. An underpass is simply a short tunnel.

Tunnel and underpass: the Tunnel panel

Open it with View → Tunnel, with the Open Tunnel button over the mono view, or with the shortcut Tunnel panel (no default key). It needs a project but no work box: a tunnel is not enclosed in a box, it develops along an axis, and its drawings are by chainage. Work down the five steps in order.

1. 1. Clean the cloud. Hide what you do not want — vegetation, vehicles, people — by drawing a closed boundary in the view; the points inside it go. This is display only, nothing is deleted. You measure on the cleaned cloud, which is why it comes first.
2. 2. Trace the axis. Draw the centreline in plan. Switch to the plan view first (T): in an oblique view a click lands where the ray meets the first surface, not where you meant it. Click along the soffit from the start to the end of the stretch — two clicks make a line, three or more a smooth curve — then press Done (Cancel abandons it). This is the alignment: you decide where it runs and the program only assigns its height. Draw the centreline again replaces it.
3. 3. Measure. Find the alignment and measure (Measure again once it is done) works along the axis section by section. When it ends the panel reports the number of sections and the length, the share of the surface actually resolved, the median radius, the number of sectors out of tolerance with the deviation (rms and 95th percentile, in millimetres), whether the alignment is a circular arc (with its radius) or straight, and the skew of the portals at the entrance and exit. Tick Show alignment and sections on the cloud to see them on the cloud. The measurement belongs to the project and is read back when you reopen it.
4. 4. Into the drawing. Axis and sections in CAD draws the alignment on a layer named ASSE-date and every measured cross-section as a real 3D polyline on SEZIONI-date, so you can snap to them and measure against them. These are the same sections as the image sheets, with no limit on their number. Each press replaces everything the tunnel engine drew before. Clear the previous drawing removes the temporary sections and those CAD entities; the measurement stays and the sections come back with Show. The panel refuses to draw if the measurement belongs to another project.
5. 5. Deliverables. A table of the drawings with their Resolution (editable) and State. Generate all the deliverables produces them in the project folder elaborati/galleria; Export them all… copies the generated ones to a folder you choose, with a short referto.txt saying which measurement they come from. Greyscale images are written as bitmaps, vector drawings are copied as they are. The Drawings tab lists what exists, with a preview you can open full size or in the image viewer.

The tunnel drawings

For Structure seen from above the program rereads the corridor and, for each pixel, takes the highest point of that chainage and offset, with its colour (greys of intensity where the cloud has no colour). The picture is saved as opera-vista-dall-alto.png in the same folder.

When a portal elevation is not measurable. The face is found where the crown ends, band by band across the opening, and a straight line is fitted through those ends. The tooltip says which of these happened, with the numbers:

• Too few crown bands — fewer than 6 half-metre bands have enough crown points. Usually the axis stops before that portal, or the cloud has no crown there.
• The edge is not a straight line — it deviates from the line by more than 0.50 m, for example because vegetation over the axis outside the structure is taken for crown.
• Beyond the end of the axis — draw the axis so that it passes that portal.
• Nothing to measure — the axis is shorter than 2 m or no points were read.

The job chain of the structure also lists the lights and point fittings, ducts and cable trays, objects to identify, and the job report.

Bridge: the Bridge panel

Choose Bridge in the Structure choice, or open the entry Deck centreline and bridge volume of the job chain; there is also an assignable shortcut Bridge panel (no default key). The panel has four steps, in this order, because you clean first and trace the deck on the cleaned cloud rather than on the tree crowns.

1. 1. Vegetation — Remove vegetation from the structure. The cloud is separated section by section: the program finds the axis as the direction in which the sections repeat most, compares each section with those a couple of metres away, keeps what persists and is smooth, and grows it over smooth cells with consistent normals. It writes ASPRS classes: 17 structure, 2 ground, 3 / 4 / 5 vegetation by height above the ground, 64 vegetation touching the structure, and leaves noise (7, 18) as it is. Classes 3, 4 and 5 are hidden when it ends; the class list in the Display panel shows them again. It reads the whole cloud and needs no centreline. A progress cover shows the phases. If a road runs right beside the bridge, check the result: the sections may follow the road.
2. 2. Deck centreline. Click along the middle of the deck in the plan view, on the track bed, then Done. The click height is kept too, so it is the height of the deck. For a railway bridge the program looks for the rails on the deck by itself, deduces the gauge, and fills the centreline from them (status line: Centreline from the rails (gauge N mm)); it stops at the bridge, 5 m beyond each end. If there are no rails it says so (No rails: axis from the sections; draw the centreline) and you draw it.
3. 3. Volume and drawings — Build the volume from the centreline. The centreline gives the axis and the deck height; the program measures the footprint of the structure, tightens the volume on it, and offers the views. If it cannot find the deck at the centreline it tells you.
4. 4. Drawings — Make the bridge drawings. See below.

The five views of the bridge

Each view isolates the structure: vegetation is left out, nothing far above the deck is drawn, and in the view from below the ground under the structure is removed too. Hidden classes stay out of every view. The five views are:

• Plan (from above) and plan from below;
• two elevations, one from each side;
• the cross section at mid-span.

The preset view buttons (front, back, top, bottom) frame the drawing that looks from that side, whichever drawing is selected in the panel. When no drawing looks that way, the computed view is kept.

Make the bridge drawings produces all five in one go, in colour, at the print resolution of 1:100 and 300 dpi, one after the other (each looking in its own direction, not that of the current view). It then opens a new CAD sheet (drawn in metres) with the images at their true size, stacked from the top left inside the paper frame, and writes a DXF (2000) in metres with the images, in the project's exports folder, as ponte_ followed by the sheet name. The status line gives the number of views, the scale and the file. It asks you to build the volume first if there is none, and it does not start a second run while one is going.

Sections from the terrain

For ground sections along a road or railway axis — not through a structure — see 10.5.3 Cross sections along an axis (Terrain → Cross sections…: layer ASSE in, layer SEZIONI, sezioni.csv and sezioni.xlsx out).

15.16 Mobile mapping: from the run to the inventory

A mobile mapping job starts from a run — the cloud, the photographs taken every few metres and the path of the vehicle — and ends in a delivery: a drawing and a list of the objects found along the road, each with its class, its position and its attributes. This section follows that path: how the run comes in, what the three views of the bottom panel are for, how the inventory is kept, and the searches that find power lines, road signs, carriageway edges, kerbs and manholes. How the views themselves behave (photograph and plan, vehicle view, drawable vehicle view) is in Mobile views; this section does not repeat it.

AI credits. Two of the steps below, the AI check of the road signs and the AI check of the manholes, send photographs to the AI service and use the prepaid AI credits of your licence. The balance is on the right of the status bar, and Help → AI Credits... shows it with the recent use. When the credits run out a check stops and keeps what it had already done; everything else in this section works without credits.

15.16.1 Importing a mobile run

File → Import mobile mapping run... reads the record folder — the one that holds Camera System and its numbered siblings — into the project. The photographs stay where they are; the project remembers where. A project with a run gets the mobile views and the bottom-panel views described below.

How a delivery is laid out is not fixed in the program: it is described by an import profile, a small text file with the extension .mmprofile. The profile says which folder holds the stitched panorama, how the ring cameras are named, and which columns of the calibration and pose files mean what. The program ships with the profile for the RIEGL VMY Camera System delivery (assets/profiles/riegl-vmy.mmprofile, in the installation folder). To read a delivery with a different layout, a profile for it is added without a new version of the program: put a copy in %APPDATA%\StereoCloud3D\profiles, where the program looks first, and then in the profiles folder beside the program.

• The program picks the profile by itself: the one that recognises the layout of the folder you chose. There is nothing to select.
• If no profile recognises the folder, or more than one does, the import stops with a message that names the folder and says why. It never guesses a layout.
• A profile file that is malformed is ignored, and the log says which and why.
• The station catalogue records which profile read it. Runs imported before profiles existed open as before.

Some deliveries arrive with a cloud that has no colour, only reflectance. See 15.16.8.

15.16.2 The three views of a mobile job

In a project with a run the bottom panel has one button per task. View → Bottom panel opens it. The run strip — station, playback, timeline, speed, View stands at the vehicle, picture follows — is always at the top of the panel, whichever view is in front.

In the Deliverable view each product has a State (not started, made, not made yet, or the number confirmed and to review), a Quantity and a Date. Show on a class row selects in the drawing the objects of that class and frames them; on a table row it opens the project folder with the table selected. If none of the objects is in the drawing any more, the status bar says so.

In every list of this section (supports, spans, signs, edges, covers...) selecting a row brings every view to the object: the panorama of the station that photographed it, the single camera, the vehicle view turned towards it, and the plan. The Up and Down arrow keys step through the rows. A station chosen this way is never refused for being far from the vehicle.

15.16.3 The inventory

Every object the job surveys — a support, a span, a sign, a manhole, a kerb — is a record in the project's inventory, the file inventory/archivio.gpkg in the project folder. It is a standard GeoPackage in world coordinates, so any GIS opens it.

The survey specification

What the job has to find is written in a survey specification (in Italian capitolato), the file inventory/capitolato.json in the project folder. It picks, from the product's catalogue of classes, the classes the customer asks for, and for each of them:

• the code the objects carry in the delivery (a class left without a code takes the catalogue default; a real job puts the customer's codes there);
• the tolerance in metres (for instance 0.20 m for a conductor, 0.10 m for a support or a sign, 0.05 m for a manhole);
• the attributes to record and which of them are required (a support: type and height; a manhole: shape and size; a sign: sign type).

It also sets the corridor around the path of the vehicle in which objects are looked for. The catalogue covers overhead lines, vertical supports, road signs, road markings, manholes, kerbs, accesses, street-light poles and electrical cabinets. A project that has no specification receives a copy of the product's example one, Rete elettrica aerea, illuminazione, elementi a terra e segnaletica, in its inventory folder, and the status bar says so. Nobody asks you to choose a file. To use the customer's own, replace capitolato.json in that folder.

The inventory follows the drawing

What is drawn on the layers of the specification — by an automatic search, by a tool or by hand — enters the inventory by itself, a couple of seconds after the drawing stops changing; it is also done when the project opens. Nothing has to be pressed.

• Each CAD entity carries the number of its record, saved in the drawing. The drawing is otherwise unchanged by the adoption.
• A moved or reshaped entity updates its record: new position, and the geometric attributes. A confirmed record becomes edited.
• An erased entity removes its record. If an erase is undone, the entity comes back with its number and is adopted again.
• As a safeguard, if more than a third of the records (and more than five) would disappear at once, the automatic update does not delete anything and writes it in the log; the button below does it when you are sure.
• A pole is one support. The support of a power line is drawn as a vertical axis plus a ring at its foot; the ring within a metre of the foot of an axis on the same layer is the same support, not a second one.
• Signs are not touched by the erase rule: a candidate sign has no entity yet.

Adopt drawing into inventory in the Deliverable view (it is only there in a project with a mobile run; a keyboard shortcut can be assigned to it) does the same on demand and shows the outcome class by class: how many objects were Adopted, were Already in, were Not adopted, and why (an entity whose shape does not fit its class, or which carries the number of a record that no longer exists). Running it again adds only what is new. The line under the button keeps the last outcome.

15.16.4 Power lines

Reconstruction → Power lines finds overhead lines along the run: first the wires, then the supports where two spans meet. It works across trees and hedges, fans out from each support and follows parallel conductors; stacked conductors are all found, each once, and a lattice tower counts as one support. What is not a wire is left out: eaves, roof edges, vineyard rows, lines lying on a surface and short pieces beside a real wire. Every span is checked in a cross-section and along its length before it is kept. The search uses all the cores of the computer and shows a progress bar that only moves forward.

Search, correct, confirm

1. Sensitivity sets how eagerly the search steps over gaps in a wire: Cautious (gaps up to 8 m: fewer false wires, more missed behind trees), Normal (12 m: a tree across the line does not stop it) or Sensitive (15 m: finds wires deeper in the trees, with a few long false lines to remove). The choice is saved in the settings.
2. Automatic search (the button is grey with the reason in its tooltip when the project has no cloud or no run). The result is a preview on the cloud; nothing is written to the drawing yet, and a line under the buttons counts what is there.
3. Correct the preview. It is easier to remove a few than to add many, so the tools are:
   • Remove — click a wire or a support on the view; what would go turns red under the cursor.
   • Undo remove — puts back the last wire or support removed; press again to go further back.
   • Add wire — two clicks on the same wire in the cloud. The program takes the wire under the cursor, not the wall behind it, follows it both ways across trees, and keeps the stretch between your two clicks and its natural continuation.
   • Add support — one click on a pole; its centre, foot and top are measured from the cloud.
   One tool is armed at a time and lit while it is; pressing the button again puts it down, and the status bar says what the armed tool expects next.
4. Confirm writes the wires on the conductor layer and the supports on the support layer, as the drawing tools do. Discard drops the preview: nothing was written and there is nothing to undo.

What the drawing holds

Under the buttons, In the drawing lists what is already drawn, whichever way it got there (search, tool, by hand), in two lists grouped by type:

• Supports, in order along the run: name, Chainage, Side (left or right of the run), Height, Type, Material, State (to review, confirmed, rejected) and From (automatic search, drawn by hand, the drawing).
• Spans: the wires with the two supports they run Between, their Length, Sag, Line and the clearances below.

The row of the list you select is shown in every view. Look moves every view onto it again (photograph from the nearest station, vehicle view, plan). Remove erases it from the drawing after one more confirmation; the inventory follows the drawing and drops the record by itself.

Clearance of the wires

Compute clearances gives, for every span, the lowest height of the wire over the carriageway — measured from the road edges, so find them first in the Carriageway tab (15.16.6) — and over the ground, from the lowest ground point under the wire. They appear as Over road and Over ground in the span list and are written to the inventory. Lowest over road and Lowest over ground move every view to the point of the wire where the clearance is smallest.

The Ground profiles, Export the tables (supports and spans, as .csv and .xlsx, indexed on the chainage of the run) and Open the report are in the Deliverable view, with the other products.

15.16.5 Road signs

Reconstruction → Road signs goes from reflective panels in the cloud to signs in the drawing, with a review step you control. It needs a project with a cloud and a run with panoramas.

1. Search signs finds the reflective panels along the run from the reflectance of the cloud and records each as a candidate in the inventory, with the photograph of the station that sees it best. A second search adds only what is new: nothing you have already reviewed is touched.
2. AI check (n) sends the photograph of each candidate not yet checked, and of each confirmed sign still without a type, to the AI service (n says how many). For each it says what the crop is — sign, street name, back of a sign, vehicle, bin, private plate, other — picks the code from the sign catalogue and reads the value and the text on it. A sign you confirmed stays confirmed, and a sign already checked is not sent again. The photographs leave your computer and the check uses AI credits.
3. Review. The list can show To review, Ruled out by the AI (candidates it says are not signs, vehicles, bins, walls, with a confidence of 80% or more), Confirmed, Rejected or All. Choosing a row shows the crop of the panorama of the station that sees the sign best (the run moves there) and, beside it, the drawing of the chosen type, to compare. The Type field is searchable by code or by name. Confirm records it as a sign and shows the next one to review; Reject records that it is not a sign, with the same next step. Rejected candidates stay in the archive, so a new search does not propose them again; from the Rejected list one can still be confirmed. In Ruled out by the AI, Reject all of these turns them all down in one go, after you have looked through them.
4. Draw confirmed (n) draws every confirmed sign again from the inventory: the panel as it stands, and in plan a circle at the foot of the pole, a stroke towards where it faces and a label with the code and the value (a question mark while the type is missing). A street name plate shows the street name. Rejected signs are taken out of the drawing, and the drawing is saved.

The type of a sign is a code of the catalogue. The Italian catalogue lists all 373 figures of DPR 495/1992 by code and name, with the kind of value each one carries (a speed, a weight, a distance...). Not every class has a drawing: artwork whose licence does not allow commercial use was left out, and those classes remain, without a picture.

15.16.6 The carriageway

Reconstruction → Carriageway traces what the road is made of. It has three searches, each with its own list, and the same way of working: the search shows a draft on the cloud, you look through the list and correct it, and only Write to the drawing puts it in the drawing. Discard throws the draft away; once a draft exists the search button reads Search again. All three need a project with a cloud and a mobile run.

Find the edges

Cuts a cross-section every half metre along the run and, from under the vehicle, follows the asphalt out to each side until it ends: kerb, grass, ditch. It steps over painted lines. The result is drawn on the cloud, green where it was measured and orange where it was deduced, and the line of text under the buttons gives the length of the run, the share measured on the left and on the right, and the width of the carriageway. Three lists:

• Edges — the edge of each side, cut at the accesses: one row per piece, with how much of it was measured.
• Accesses — where the edge opens because the asphalt goes on outwards: crossings, driveways, lay-bys. Unpaved ones are not found.
• To review — stretches where the edge was not measured and is not an access: a parked car, a hidden stretch, the road losing its edge. The line there simply joins the measured ones on either side.

On a row, Look moves every view to it, and Leave out keeps it out of the drawing (Take back undoes that). Write to the drawing puts the edges, cut at the accesses, and the accesses on the layers of their inventory classes; rows you left out are not written. Where the survey specification has these classes, they enter the inventory by themselves.

Without a draft, the lists read what is in the drawing, with Look and Remove (it asks once more). While a draft is open they are headed Found - not yet in the drawing.

Find kerbs

Follows the face of the kerb as one continuous line, seen from above as a ridge in the ground, instead of deciding cross-section by cross-section. It looks for a step between two flat surfaces (the road and the top of the kerb, both smooth to within 5 mm, which leaves grass verges out) and follows it, turning corners around traffic islands. Because the line follows the face, it replaces the short pieces that the cross-sections alone would draw at a zigzag; sloped kerbs and the places where the road ends at a wall are still found from the cross-sections. A real narrow kerb that the line does not reach does not appear.

• Lowered kerbs — where the step is under 5 cm (a ramp, a driveway) the kerb is followed through it and drawn in yellow on the cloud.
• Behind parked cars the kerb cannot be seen. It is continued between the two measured ends and marked as not measured: deduced when the top edge of the kerb or the back of the pavement was seen over the car, hypothesised when nothing was seen and the curve just joins the two ends. Look at these before confirming: behind a car there may be a driveway. They are drawn dashed on the cloud.
• Two lists: Kerbs, grouped by kind (Kerb, Kerb, partly behind parked cars, Narrow curb, Sloped kerb, Road edge at a wall), and Kerb gaps, where a kerb stops and starts again, with the reason: a flush opening (driveway, crossing), a lowered kerb, a ramp, an obstacle, a car in front, no points.

Write to the drawing puts the measured kerbs on the kerb layer, the stretches not measured on a layer of their own, and the walls at the road edge on theirs, so the three can be told apart and handled separately. Rows left out are not written.

Find manholes and gullies

Looks for covers only (painted road markings are not traced here). Find manholes and gullies grows the road surface from the path of the vehicle up to the kerbs and finds the patches darker than the asphalt in the laser, with the real size of a cover (about 20 cm to 1.4 m) and a compact shape. They appear in the Manholes list as dark patches.

Then AI check (n) checks them. For each object two images go to the AI service: the laser seen from above with the object outlined in red, which says where it is, and the photograph of the same spot from the station that sees it best, which says what it is: a manhole, a gully grating, a small lid, or other (a patch, a crack, a shadow). Objects with more than a third of their area without points are noise — a car stood there — and are not sent. Answers below 70% stay to review, and an unsure answer is looked at again in a photograph. The same step also looks for covers the laser missed: the ground along both sides of the road, pavements and verges included, is seen from above out of the photographs of several stations, and each cover found is confirmed from a second station. Measured on three stretches of an urban run this found about 44 of 46 covers, against 14 from the laser alone. The check uses AI credits; if they run out the status bar says so at once.

The photograph window. Selecting a row of the Manholes list opens Photos of the object: the object seen from up to four stations around it, with a circle on the point. Below, you decide: manhole, gully grating, small lid or Not a cover (the object stays in the list, left out of the drawing, and can be changed again). After each decision the next object not yet decided is selected: forward along the run in the same group, then from the start of the group, then the other groups. If no station sees the object from 3 to 15 m past the vehicle the window says so; with no photographs at all the cloud's multiview opens on the point instead. The kind can also be changed in the list: the row moves to its list and goes back into the drawing.

Write to the drawing puts each manhole and gully on the layer of its inventory class: with its own outline when the laser reads the shape clearly (round or rectangular), as a placeholder point with its name otherwise. What is not a cover, and rows left out, are not written.

15.16.7 Lists grouped by type

Every list of detected objects — signs, supports, spans, covers, edges, accesses, kerbs, stretches to review — is divided into groups of one type. A group opens and closes with a click; the group of the selected row opens by itself, so the object you selected from a view is never hidden in a closed group. Rows without a type go in the group type not set. Selecting a row brings every view to the object (see 15.16.2).

15.16.8 A cloud without colour

If the cloud of a mobile run arrived without colour — the colour fields are empty, or all black — the job's data column says so, and when the project has the photographs of the run it adds the button Colour the cloud from the photos. It is the same command as Cloud → Colour the cloud from the run... (a keyboard shortcut can be assigned to it): it paints every point with the colour of the photographs the vehicle took, rewrites the cloud in place, and can take several minutes. Nothing is changed until you press it.

16. 3D Inspector window

The internal 3D Inspector renderer (Inspector3DView) remains in the codebase for vector/scene previews, but the floating View → 3D Inspector window is not available in the current release. (Ctrl+Shift+I now toggles imagery underlays.) Use the main viewport, trackball orbit, and oblique stereo views for oriented inspection instead.

17. UI magnifier (loupe)

Because some menu text is small, an optional loupe magnifies the area under the pointer while it hovers the user interface. Toggle it with View → UI magnifier or Ctrl+Shift+M, and tune it in the Inspector:

• Zoom factor — 1.5× to 6× (default 2×).
• Lens size — 80 to 600 px (default 220).

18. Keyboard shortcuts

Shortcuts are fully customizable in CAD → Keyboard Shortcuts…, where you can rebind any managed command, check for conflicts, and Reset all to defaults. The panel also lists the typed CAD commands for reference. Default bindings:

In addition, the default feature macros (P, B, C, Shift+G, V, D, W, R, H, S, X, L, A, O, N) fire when no managed command is active. Managed shortcuts always take precedence over macros.

19. Command line

The Command Line panel accepts typed commands and shows operation results. Type HELP (or ?) to list the available verbs. Useful application verbs include:

NEWOPENSAVE SAVEASIMPORTEXPORT FITTOPPM IMAGERY / ECW SNAPXYZ / SNAPXY / SNAPZ SH / QUOTEPROPSGROUP GROUPSLA (Layer)LAO (Layer Order) CLEAR / CLSNAV VOLUMI / VOLUMES / VOL AICREDITS