What you will learn
- What a point cloud contains, and what classification adds
- How ground models are extracted from classified clouds
- What quality checks matter before a deliverable leaves the office
- How large clouds are stored, tiled, and shared without choking
- Where LiDAR fits alongside conventional survey measurements
What the project is
LiDAR scanners measure a site with laser pulses, producing a dense cloud of points with coordinates and intensity. Ground-based and airborne scanners both produce clouds, and both need the same downstream work: removing noise, classifying points into ground, vegetation, buildings, and other classes, and deriving surfaces and features from the classified result.
The platform covers that downstream path. It ingests scans, registers them into a common coordinate system, classifies points, builds ground models, and exports deliverables. It is a processing and delivery system, not a scanner, and it works with data from scanners the survey team already owns or hires.
Core features
Ingest and registration come first: scans arrive in different formats and coordinate systems, and the platform must bring them together with a recorded transformation and a fit quality report. Classification follows, with ground extraction as the most important class, because most deliverables depend on the ground model. Tiled viewing lets users inspect a large cloud in the browser without downloading gigabytes.
- Multi-format ingest with coordinate transformation and fit reporting
- Point classification: ground, vegetation, buildings, noise, and custom classes
- Ground model and contour generation from classified ground points
- Tiled browser viewing for large clouds without full downloads
- Quality reports: density, coverage gaps, and registration residuals
- Export to surface models, CAD, and GIS formats
How it is built
Start with ingest, registration, and ground classification for one scanner type, and prove the ground model against conventional survey checks on a known site. Viewer and export features follow, because a perfect classification that nobody can inspect or deliver is not a product.
- Define the deliverable set with a design or survey team.
- Build ingest and registration with fit quality reporting.
- Build classification with ground extraction, tested against surveyed check points.
- Add tiled viewing, quality reports, and the export set.
- Add project management, storage tiers, and client delivery.
Technology considerations
Scale is the deciding technical question. A single site can produce tens of millions of points, so storage, tiling, and processing must handle that volume without forcing every user to download the whole cloud. Cloud processing with tiled storage is the usual answer, but costs scale with data volume, and the pricing model should reflect that plainly.
Classification quality needs ground truth. Automated classification improves with training data from the team's own sites, and the platform should let surveyors correct classes and feed those corrections back, rather than treating the first automatic pass as final.
Challenges to plan for
Vegetation is the persistent problem: dense canopy hides the ground, and the ground model under trees is an interpolation, not a measurement. The platform should show point density and coverage gaps alongside the model, so users can see where the ground is measured and where it is inferred.
- Ground hidden under vegetation, producing inferred rather than measured surfaces
- Registration errors between scans that shift parts of the cloud
- Storage and processing costs that grow with every project retained
- Deliverables that look precise but carry classification errors
Who it is for, and its limits
This platform fits survey firms that own or hire LiDAR scanners, engineering firms processing scan data for design, and agencies managing corridor or terrain surveys. It is most valuable where scan volume is regular, not occasional.
It does not replace conventional survey checks. Ground models should be verified against surveyed control and check points, and areas with poor coverage should be flagged in the deliverable. Costs and timelines vary with data volume and processing approach, so confirm with a quote before budgeting.
Proving the ground model before delivery
Before a ground model leaves the office, it should be checked against independent survey measurements on the same site. Pick check points across the terrain, including flat areas, slopes, and vegetated ground, and compare the model against them. The differences tell the team where the model is trustworthy and where it needs conventional survey infill, and that comparison belongs in the deliverable notes, not in a separate file nobody reads.
Agree the check regime with the client at the start of the project, including how many check points, where they sit, and what tolerance applies. A model delivered with its check report is a survey product; a model delivered without one is a picture that looks precise. Clients who understand that distinction will trust the next delivery more, and will ask fewer questions about the one they already received.
Frequently asked questions
What is a point cloud?
A dense set of measured points with coordinates, produced by a laser scanner. Each point records where a laser pulse hit a surface, and together the points describe the site in three dimensions.
What does classification do?
It labels points by what they hit: ground, vegetation, buildings, noise, and other classes. Classification is what turns a raw cloud into a usable ground model and feature set.
Why does vegetation cause problems?
Dense canopy blocks laser pulses from reaching the ground, so the ground model under trees is interpolated from nearby points rather than directly measured. Coverage maps should show where that happens.
How is accuracy checked?
By comparing the processed ground model against independently surveyed check points and control, and by reviewing registration residuals between scans.
Do users need to download the whole cloud?
No. Tiled browser viewing lets users inspect a large cloud without downloading it, and full downloads are available in standard formats when needed.