Identifying ground

This exercise uses PDAL to classify ground returns using the Simple Morphological Filter (SMRF) technique.

Note

This excerise is an adaptation of the pcl_ground tutorial on the PDAL website by Brad Chambers. You can find more detail and example invocations there.

Exercise

The primary input for Digital Terrain Model generation is a point cloud with ground vs. not-ground classifications. In this example, we will use an algorithm provided by PDAL, the Simple Morphological Filter technique to generate a ground surface.

See also

You can read more about the specifics of the SMRF algorithm from [Pingle2013]_

Command

Invoke the following command, substituting accordingly, in your Conda Shell:

1
2
3
4
pdal translate ./exercises/analysis/ground/CSite1_orig-utm.laz \
-o ./exercises/analysis/ground/ground.laz \
smrf \
-v 4
1
2
3
4
pdal translate ./exercises/analysis/ground/CSite1_orig-utm.laz ^
-o ./exercises/analysis/ground/ground.laz ^
smrf ^
-v 4
../../../../_images/ground-run-command.png

As we can see, the algorithm does a great job of discriminating the points, but there’s a few issues.

../../../../_images/ground-classified-included.png

There’s noise underneath the main surface that will cause us trouble when we generate a terrain surface.

../../../../_images/ground-classified-included-side.png

Filtering

We do not yet have a satisfactory surface for generating a DTM. When we visualize the output of this ground operation, we notice there’s still some noise. We can stack the call to SMRF with a call to a the filters.outlier technique we learned about in denoising.

  1. Let us start by removing the non-ground data to just view the ground data:
1
2
3
4
5
6
pdal translate \
./exercises/analysis/ground/CSite1_orig-utm.laz \
-o ./exercises/analysis/ground/ground.laz \
smrf range \
--filters.range.limits="Classification[2:2]" \
-v 4
1
2
3
4
5
6
pdal translate ^
./exercises/analysis/ground/CSite1_orig-utm.laz ^
-o ./exercises/analysis/ground/ground.laz ^
smrf range ^
--filters.range.limits="Classification[2:2]" ^
-v 4
../../../../_images/ground-ground-only-view.png

2. Now we will instead use the translate command to stack the filters.outlier and filters.smrf stages:

1
2
3
4
5
6
7
8
9
pdal translate ./exercises/analysis/ground/CSite1_orig-utm.laz \
-o ./exercises/analysis/ground/denoised-ground-only.laz \
outlier smrf range  \
--filters.outlier.method="statistical" \
--filters.outlier.mean_k=8 --filters.outlier.multiplier=3.0 \
--filters.smrf.ignore="Classification[7:7]"  \
--filters.range.limits="Classification[2:2]" \
--writers.las.compression=true \
--verbose 4
1
2
3
4
5
6
7
8
9
pdal translate ./exercises/analysis/ground/CSite1_orig-utm.laz ^
-o ./exercises/analysis/ground/denoised-ground-only.laz ^
outlier smrf range  ^
--filters.outlier.method="statistical" ^
--filters.outlier.mean_k=8 --filters.outlier.multiplier=3.0 ^
--filters.smrf.ignore="Classification[7:7]"  ^
--filters.range.limits="Classification[2:2]" ^
--writers.las.compression=true ^
--verbose 4

In this invocation, we have more control over the process. First the outlier filter merely classifies outliers with a Classification value of 7. These outliers are then ignored during SMRF processing with the ignore option. Finally, we add a range filter to extract only the ground returns (i.e., Classification value of 2).

The result is a more accurate representation of the ground returns.

../../../../_images/ground-filtered.png