Filtering & Classification
This is where a raw point cloud becomes usable. Three jobs, in order: remove what should not be there, classify what is left, and normalise the heights so each point carries its height above the ground rather than its altitude above sea level.
Most of these tools read a LAS or LAZ file and write another one, which is downloaded to your computer. Chain them by feeding each output into the next tool.
Start here
Section titled “Start here”Cleaning
Section titled “Cleaning”Filter Lidar Noise removes the standard noise classes, 7 for low noise and 18 for high noise, that most providers already flag. The first thing to run.
Lidar Remove Outliers goes further and detects points that disagree with their neighbours, which catches birds, low flying artefacts and measurement blunders that carry no noise flag.
Filter Lidar Scan Angles removes the returns collected at a grazing angle, at the outer edge of each swath. They are the least accurate points in the survey, and dropping them measurably improves a terrain model.
Remove Duplicates drops points sharing the same position, which happens where flight lines overlap.
Classifying
Section titled “Classifying”If Lidar Info told you the cloud has no ground class, you need one before anything else works.
Classify Lidar does the general job, separating ground, vegetation and buildings from the local geometry of the cloud.
Improved Ground Point Filter and Lidar Segmentation Based Filter focus on the harder and more important question, which points are ground. The second grows connected components from locally flat regions, which is robust in vegetated terrain.
Classify Buildings In Lidar takes a different route: give it building footprints as polygons and it assigns class 6 to every point inside them. When you have a cadastral layer, this beats any geometric method.
Normalising
Section titled “Normalising”The step that unlocks everything about vegetation and buildings.
Normalize Lidar subtracts a terrain raster from the point elevations, so a point at 3 metres is 3 metres above the ground, wherever the ground happens to be.
Height Above Ground does the same without needing a raster, by comparing each point with the nearest lower ground point in the cloud itself.
Once normalised, a canopy height model is simply the maximum height per cell, and tree detection, canopy metrics and building heights all become straightforward.
Selecting
Section titled “Selecting”Clip Lidar To Polygon keeps only the points inside a polygon layer, which is how you cut a study area out of a tile. Erase Polygon From Lidar does the opposite.
Filter Lidar is the general purpose one: a boolean expression over point attributes such as class, elevation, return number or scan angle.
All the tools
Section titled “All the tools”| Tool | What it does |
|---|---|
| Filter Lidar Noise | Removes the ASPRS noise classes, 7 and 18. |
| Lidar Remove Outliers | Detects and removes points that disagree with their neighbourhood. |
| Filter Lidar Scan Angles | Removes returns beyond a scan angle threshold. |
| Remove Duplicates | Removes points sharing the same position. |
| Filter Lidar Classes | Removes points of the classes you name. |
| Filter Lidar | Removes points using a boolean expression over their attributes. |
| Filter Lidar By Percentile | Keeps one point per cell, at a chosen elevation percentile. |
| Filter Lidar By Reference Surface | Keeps points above, below or within a threshold of a reference surface. |
| Lidar Elevation Slice | Extracts or reclassifies the points within an elevation band. |
| Classify Lidar | Automatic classification into ground, vegetation and buildings from local geometry. |
| Improved Ground Point Filter | Multi stage ground point extraction. |
| Lidar Segmentation Based Filter | Ground filtering by growing components from locally flat regions. |
| Classify Buildings In Lidar | Assigns the building class to every point inside footprint polygons. |
| Classify Overlap Points | Detects and flags or removes points in flight line overlaps. |
| Lidar Classify Subset | Transfers a classification from an auxiliary cloud onto the base cloud. |
| Lidar Segmentation | Partitions the cloud into connected planar segments, storing segment ids. |
| Individual Tree Segmentation | Groups vegetation points into individual tree crowns. |
| Normalize Lidar | Subtracts a terrain raster, converting elevations into heights above ground. |
| Height Above Ground | Computes height above the nearest ground point, without needing a raster. |
| Modify Lidar | Applies transformations to point attributes, such as an elevation offset or a reclassification. |
| Clip Lidar To Polygon | Keeps the points inside a polygon layer. |
| Erase Polygon From Lidar | Removes the points inside a polygon layer. |
Order matters. Clean before classifying, classify before normalising, normalise before measuring vegetation. Each step assumes the previous one is done.
Keep the intermediate files. Every tool writes a new file. Name them by step, so that when a result looks wrong you can go back one stage rather than start over.
Check what you removed. Run Lidar Info again after a heavy filtering pass. A cloud that lost half its points is telling you the threshold was too aggressive.