Ground Classification Algorithm

[dpitman]

I came across this algorithm for filtering non ground points using CloudCompare. Cloth Simulation Filter (CSF)
Has anyone any experience with it?  It's very interesting and looks like it results in a nice smooth surface. For those in the know, how does it compare with how MS approaches this?

thanks,
Dave

https://www.cloudcompare.org/doc/wiki/index.php?title=CSF_(plugin)

[c-r-o-n-o-s]

You cannot compare them.
CloudCompare can use the other approach, that the terrain is seen from "below".
The results are also only good if the parameters are correct.

[dpitman]

You cannot compare them.
CloudCompare can use the other approach, that the terrain is seen from "below".
The results are also only good if the parameters are correct.

Perhaps; Please explain why.

Looking at the simple illustration of how the algorithm selects the surface seems to find the more accurate "ground" between dense trees, for example.  The points of ground that are between trees creates the "bottom" and that is what I want to find for a DEM.  MS amazes me when it classifies ground points, but it always leaves a little bit at the bottom of what it removes.  I'm not saying I tried every different setting, but I've tried a lot.  And in areas where dense vegetation is removed, rather that using the low points left from around the removed vegetation from which to define the surface, there is usually what looks like a triangle mesh garden.  While the smooth surface that would result from the Cloth Simulation approach nor the MS approach will be anywhere near as accurate as the bare earth areas, I think it is much less "noisy".

I'm not sure what you mean by "if the parameters are correct"?

[dpitman]

In the screen capture, you can see what remains after auto classification.  Of course this can be "smoothed".  What would be preferable is to specify various points of the surface, the low points, and have everything else removed and filled by extrapolation.  I think that would result in a better representation than "smoothing" which has not control.  Wouldn't the Cloth surface approach result in a closer representation of the approximated bare earth surface?