DTM Extraction Settings

Based on comparison with other photogrammetry software offerings, I’ve been somewhat disappointed in how thorough DroneDeploy’s algorithm is at filtering vegetation, structures and other non-bare ground elements from the point cloud and resulting DTM. By comparison, Metashape offers very granular control over the geometry parameters for the filter. However this can be time-consuming to experiment with settings by trial and error that work best given each site’s potentially unique features. On the other hand, other software can be too aggressive and filter out abrupt ground features that should have not been filtered.

My request is that the development team do some experimenting with different settings and test data and possibly develop a “mild”, “standard” and “aggresive” algorithm set and let the user select the aggressiveness setting of the filter.

Another potential option is to add a relative simply point cloud editing feature, that allows the user to specify a color range to filter from the point cloud. In many cases the color of the bare ground is similar where as non-ground features such as bushes, trees, vehicles and structures are of a significantly different color. The current algorithm does a so-so job of filtering out these features (obviously no machine algo will be better than a manual human operation) but giving the user more options to improve the filtering of non-bare earth features would be highly valuable for the Surveying/Mapping application sector.

2 Likes

Creating DTM’s is very tricky. Across all of the photogrammetry solutions that I have used I have never seen one that creates a truly accurate DTM. Like you said they either don’t get rid of enough, are very time consuming or are way too aggressive. This is why we chose to do all of our DTM’ing in Carlson Precision 3D Topo. It does a true filtering method using a cell, window and elevation delta. It’s also very good at getting rid of outliers once you have gotten rid of the initial unneeded points. The other thing that’s really nice is that it saves those points to another layer so that they are always present in the Carlson file as well as the fact that you can bring in other point clouds for merging.

DroneDeploy’s algorithm has gotten better over time but they really struggle with things that are larger than the average car or tree. It has a really hard time recognizing the difference between ground and stockpiles which is something I don’t think you can do in an automated process across the vast types of maps that they have to deal with. It’s easy to create a setting that works for you but when you try to create something that works for everything that usually doesn’t work very well.