PilotLedger emblem PilotLedger
← Back to Blog
PL

What Is Point Cloud Classification? Ground, Trees and Noise

Point cloud classification is the step that turns a cloud of millions of unlabeled points into something you can build a deliverable from. Each point gets a label: ground, vegetation, building, water, noise and so on. Those labels decide which points go into a bare-earth surface, which ones get filtered out of a volume, and which ones a client's engineer will trust.

This guide explains what the common classes mean, why classification matters for DTMs, contours and quantities, and how to review a classified cloud before it leaves your hands.

What point cloud classification means

A raw point cloud is just coordinates, often with color or intensity attached. It does not know the difference between a roof, a tree canopy and the ground under it. Classification assigns each point to a category so you can select, hide or exclude points by what they represent.

The labels themselves are standardized. The ASPRS LAS specification defines numbered class codes that most mapping and GIS software understand, so a cloud classified in one tool can be read in another. Common ones include:

Projects can also use custom classes for things like equipment, stockpiles or fences, as long as everyone downstream knows what the codes mean.

Why classification matters for your deliverables

Most of the products clients pay for depend on the ground class being right.

Getting ground right is usually the whole game. The other classes mostly exist so you can get them out of the way.

Manual vs automatic classification

Classification can be done by hand, by an automatic routine, or by a mix of both.

Automatic classification uses rules or trained models to label points in bulk. It is fast on large sites, and it handles open, gently sloped ground well. It struggles where the rules break down: steep banks, dense brush, low walls, stockpile toes and anything that looks like ground but is not.

Manual classification means selecting points and assigning classes yourself. It is slower, but it is how you fix the places an automatic pass got wrong, and on small or messy sites it is often the only reliable option.

In practice, most commercial work uses an automatic pass followed by manual cleanup in the areas that matter for the deliverable. Whichever route you take, the responsibility for the result stays with whoever delivers it.

Photogrammetry clouds vs LiDAR clouds

Where the cloud came from changes what classification can do.

A photogrammetry cloud is built from overlapping photos, so it only contains surfaces the camera could see. Under dense canopy there may be no ground points at all, and no amount of classification can recover terrain that was never captured. Open sites, pads and stockpile yards classify well. Wooded lots often do not.

A LiDAR cloud is built from laser returns, and some pulses reach the ground through gaps in vegetation. That is why LiDAR is the usual choice when bare earth under trees is the deliverable. It still needs classification and review, but it starts with more ground to work with.

Common classification mistakes

Review the classification before delivery

A class label does not establish survey accuracy. Before a classified cloud or a surface built from it goes out, inspect it against the project's control and the deliverable requirements. Pay particular attention to equipment, canopy, roofs, water and gaps in capture coverage. Note in your delivery what was classified, how, and what was left alone.

If the job needs a stated accuracy or a sealed survey product, that comes from control, checkpoints and a licensed professional, not from a classification pass.

Where SkyView fits

SkyView Desktop builds RGB orthomosaics and DSM and DTM surfaces on your own Windows PC, with no per-map processing fee, and delivers finished maps to clients through private Pilot Ledger projects. For more on the surfaces that depend on good ground points, read about DSM and DTM surfaces and cut-and-fill volumes.

SkyView is available now. See plans at pilotledger.com/#pricing.