CloudCompare forum

I am trying to detect a specific physical feature (not based on RGB) from the dense point cloud. For example, across a landscape with many ground pits (holes), I would like to detect ground pits (holes) with specific shapes. I was thinking of a supervised way where I delineate specific shapes (3D) I want and the program finds (or classify) the area with similar/identical features. I am aware that the CloudCompare has plugins like CSF and CANUPO but I am not sure their algorithm is based on physical distances (and features) but rather an RGB...

Actually Canupo is probably your best chance as it is based on the shape and not on color. Pay attention that the scale of things has a big importance though. If your holes have more or less the same scale then it should be ok. Of course you can also use several classifiers.