source code found at -
The tweaks..
Parameters: (these settings were optimized with a Poisson filter at Radius = 1. I would also suggest setting hres to 1)
s= 20
k= 0
n= 0.1
b= -0.2
I also stacked a default PMF filter and a cell to write out the final ground points.
*note - It was critical for my tequitor's processing time to keep the point density around 1pts/m^2 (Yes, my CPU is a burrito).
If you have a Debian or Ubuntu environment, there's a setup script that stands a decent chance of getting you going with a local Python environment.
Note: using Conda instead of python-venv, because the PDAL developers provide their latest packages directly to the Conda Forge repos, but putting the environment in a subdirectory within the project directory, which maybe keeps things cleaner.
Clone this repo, cd into it, and type
script/setup.sh
If you don't have a Conda environment, it's going to try to install one, which might require a sudo password.
Once it's done, you need to activate the environment with
conda activate ./envs
At which point you can probably start filtering point cloud files in .laz format with
python3 app/dtm_filter.py <inputfile>
If you know what the various parameters mean, you can set them like so:
python3 app/dtm_filter.py <inputfile> -S 10 -k 0.0 -n 0.5 -b -0.2 -hres 1.0
(those are actually the default values, which will be used if you don't specify any parameters at all).
So you have a filtered point cloud containing only ground points, with vegetation and buildings removed! Great! But your friendly neighborhood hydrological modeler probably wants a GeoTIFF raster file. WebODM has a utility that handles conversion from a point cloud to a DEM raster, and you can run it standalone.
Clone the WebODM repo and follow the setup instructions in the WebODM Readme. Once it's set up, cd into the WebODM directory and issue the following command use this command documented here in the WebODM repo:
docker run -ti --rm -v 'PATH/TO/DIRECTORY/CONTAINING/INPUTFILE:/input --entrypoint /code/contrib/pc2dem/pc2dem.py opendronemap/odm /input/INPUTFILENAME.las --type dtm
Then you probably want to clip it with the same boundary as the dsm and dtm from OpenDroneMap. I'll automate this later with pygdal, but for now I'm just doing it in QGIS with the r.mask.vect tool using the odm_georeferencing/odm_georeferenced_model.bounds.gpkg as a mask layer, and saving the resulting layer as a GeoTIFF with COMPRESSION DEFLATE.