Basic DSM output straight from WebODM.
I've completed a few projects now, and have got a pretty good understanding of how things work. But as expected, there's huge room for improvement with my results. It's been great experimenting with the different outputs from WebODM and initially I was happy to see the basic DSM/DTM type outputs. But now I understand what's going on better, I can see a lot of errors in the outputs. Here's an outline of what i've realised and how i've begun to refine my process...
WebODM is an all-in-one processor that will generate point cloud, 3D models, DTM, DSM, and Orthomosaic, and with the default settings is effectively a one-click process. I've used it like this for a few test projects and have been satisfied with the results.
But something didn't feel quite right - I knew the point clouds were a bit messy and had 'noise' around the main subject, and had been exploring ways to tidy this up (in Cloud Compare). Then I noticed that the colour grading across the DSM/DTM outputs was a bit sloppy and didn't reflect the actual terrain variations. The penny dropped - the DTM/DSM is generated from the point clouds, and the colour scales were being distorted by the false high and low points in the noisy areas of the cloud.
A bit of playing with Cloud Compare was enough to clean up the most obvious noise and make the point cloud cleaner. From there I could take the point cloud into QGIS and use its own tools to generate a fresh DSM and DTM.
This also helped me to understand how the two different surfaces were being generated. I knew about Classification and that WebODM tries (with reasonable success) to identify what each point forms part of. The DSM then includes 'everything' (trees, buildings, etc), but the DTM uses the classification to filter out just the points that relate to the ground and interpolates the gaps to infill the remaining areas.
As you can see below, the results from the cleaned point cloud are far better than the original. The cleaned version below also has an added 'hillshade' layer to add definition and clarity to the image.
I have also now realised that WebODM does have it's own settings to help automatically clean the point cloud, so I will be using these options in future as well as further tidying in Cloud Compare.
Mobirise.com