The Auto-classifier pipeline includes useful Presets for common pre- and post -classification tasks. Use them as-is, or tune them based on your LiDAR source data.
By default, the presets apply common denoising settings to these stages suitable for typical LiDAR classification tasks:
To illustrate how these options work, the following screens display partially-classified data such as that generated in a classification preview (see below for information about running classification previews).
However you can also apply these steps to unclassified data.
How the denoising presets work
Consider this partially-classified LiDAR data that has been imported, and which displays outlying points classified as type Fence:
To inspect a point's properties, choose the Point Cloud tool on the toolbar, and select a point:
In this LiDAR capture, some of these points may be part of a real Fence, some may be noise. Depending on your data, the pre- and post- processing presets can help minimise or eliminate the noise.
To see the effect of the Presets, expand the Advanced options section, and then expand the Denoise step, in a stage in the Pipeline, and then select a different preset. You'll notice that the denoise parameters values will change:
Considering the Medium preset, we're interested in these two parameters:
Min Neighbour Count
Nearest Point Distance
In this example a point will be classified as noise unless there at least 4 neighboring points within a Nearest Point Distance of 2 meters
You can use the Clearance Measure tool on the toolbar to see how the outlying points in the LiDAR compare to this preset:
If you don't see the Clearance Measure tool, click the More icon (…) at the bottom of the toolbar
Select the Clearance Measure tool, click on one of the outlying Fence points, and then hover over a nearby point to see the distance between the two:
Notice in the video that there aren't at least 4 neighbouring points within 2m of the selected point. In this case, this preset will result in that point being classified as Noise.
If we select the Aggressive preset, notice that the denoise parameters change:
Using this preset, a point will be classified as Noise unless there at least 5 neighboring points within 1 meter.
In the example above and based on the measured distances, the selected point will still be classified as Noise. However this preset might now result in incorrect denoising in other parts of the data.
To demonstrate, we'll zoom into a part of the LiDAR that appears to be a valid fence-like structure, and measure the distances again:
Notice that the selected point has 5 neighboring points, but only some of them are within 1 meter away. In this case, these will be classified as Noise - so this preset will filter out potentially useful, non-noise data points.
In practice, the values you set will depend on your source LiDAR data. We recommend that you start with a lighter default preset and then tune the parameters by inspecting parts of data.