CloudCompare forum

Hello All,
Two questions, if you will indulge me...

1)I am working on a project to try to quantify the rate of erosion at a lithified footprint site in Tanzania. I have imagery from 2010 and 2017, but there are no permanent ground control points. I have tried using the point picking and the ICP alignments based on areas that seem to have the least erosion but the alignments ultimately try to split the difference between positive and negative change. Since there has been far more erosion than deposition, this skews the result. Is there a better method of alignment I should be trying? 2) I also have used the M3C2 algorithm to estimate change, but all my points come back red (insignificant change). In the attached image, you can see the change is real, but something in my parameters is not scaled correctly perhaps. Parameter and normal settings attached as well. Any help you can provide is greatly appreciated.

Thanks!

1) The ICP alignment tool will indeed try to minimize the distances between the two clouds (and you'll get therefore distances centered about 0)... apart if you set the 'overlap' parameter correctly. If you set it to 10% or 20%, then only the closest points will be used for registration.

Of course, if you rely only on the point-pair registration tool, then you'll avoid this effect completely. This tool is much more reliable when dealing with heavily modified objects.

2) Isn't the registration error already quite high? (relatively to the scales, etc.). Can you try without it?

And don't hesitate to send me your dataset (cloudcompare[at]danielgm.net), it will be easier for me to help you.

Hi Daniel,
Thanks so much for getting back to me.

I'll keep trying to align the prints manually to get an optimized alignment. In this case, do I just use the RMS from point picking as the "registration error" in M3C2?

I ran a couple of M3C2 tests and the data are all colored red for insignificant change even when I do not check "Registration error" or if I use .0000001. Perhaps the texture of the footprint itself is being interpreted as surface roughness?

I will send you my dataset directly.

Thanks!
Brian

I just realized: '1' for the 'significant change' scalar field means that all points are considered as being a 'significant change' actually ;D.

I did a quick and dirty registration (ICP with 30% of overlap - I didn't know if you had already registered the clouds). It gave me a 0.00045 RMS. And then I used the 'auto guess params' button in M3C2 (still quick & dirty) with vertical normals.

And the result seems quite consistant: For the records, the projection diameter was 0.000708, the normals one was 0.001416 and the depth was 0.020349.

Hi Daniel,
This is great news. Thanks for taking the time to run the clouds and for the clarification. I think my primary confusion came from the fact that under the "significant change" tab, all pixels were either colored red or grayed out. In Lauge's paper and on the wiki page, red is identified as non-significant change and gray means no matched points. I am thrilled to see that this analysis will work.

Thanks so much for the help!
Brian