CloudCompare forum

I'm scaling SFM models (made from old (50-110 yr old images) to modern lidar point clouds. The models are of large mountain walls and glaciers, and the goal is to identify volume change over time. Specific questions I have include:

1) How best to align models in which I expect large areas to have significant change (in the case of glaciers especially)
2) Once aligned, how best to evaluate C2C distance, because the C2C distance tool only registers positive difference, so is it better therefore to use M3C2.

I'm using the following workflow, which I came up with on my own and I am wondering if anyone has experience with this or can comment if the see flaws in my approach:
1) Import LIDAR and SFM point cloud.

2) Using SFM point cloud as "aligned" entity I do registration with LIDAR as reference, checking the box, "adjust scale." My assumptions is that CC scales the "aligned entity" TO the fixed ("reference") entity. I am aware that failure to keep this organised through the iterations will result in an altered scale for my LIDAR model.

3) Fine registration. First iteration: 1e-5 RMS difference, final overlap of 10%, "adjust scale" selected, and 50000 random sampling limit. I do 3-4 iterations increasing RMS by 1 order of magnitude (-5,-6,-7), and multiply random sampling limit by 4 each time. Normally RMS is somewhere between 0.5 and 1.5

4) Use either C2C distance algorithm and/or M3C2 to see areas of significant change.

5) Volume change calculation. I haven't go to this step yet, any tips appreciated!

If anyone has insight into my workflow, or sees major flaws, please tell me.

2) Yes, the algorithm will adjust the scale as well as the position to improve the fit between the 2 entities

3) I'm not sure that increasing the RMS difference will change the result drastically (it's just the RMS difference between 2 iterations, so 1e-5 is already very small). And for the random sampling limit, the more the better (it's just that the process takes more time).

The only interesting thing is probably the fact that at each iteration, the 10% closest points change a little bit. So maybe increasing the overlap percentage, and using a high sampling limit, would allow to achieve almost the same result in one step?

5) It depends on your clouds? But if they are dense enough, then the '2.5D Volume calculation' tool should give interesting results.

Thanks!