Subtraction Errors found in M3C2

[RiverRunner]

Hello,
I am subtracting two point clouds and am finding some extreme errors in my M3C2 subtraction where there are no points located in the original point clouds.

Here is the M3C2 difference between the two clouds. This projection is using the mesh 2.5D best fit with 1m max edge length to visualize the error.
As you can see there are the pinnacles of red and blue which are a difference of more than 5m (impossible at that spot).

Screenshot 2024-06-11 at 12.01.02 PM.png (923.48 KiB) Viewed 1314 times

As you can see from the point cloud, that there are no points in those areas to create that pinnacle. First is the baseline, the next is baseline overlaid with M3C2

Screenshot 2024-06-11 at 11.55.56 AM.png (823.32 KiB) Viewed 1314 times

Screenshot 2024-06-11 at 11.56.52 AM.png (913.99 KiB) Viewed 1314 times

My parameters for the M3C32 is as follows: I used the "guess parameters" feature and it provides a value double my point cloud density distance.

Screenshot 2024-06-11 at 12.06.42 PM.png (168.59 KiB) Viewed 1314 times

Screenshot 2024-06-11 at 12.07.07 PM.png (157.95 KiB) Viewed 1314 times

Screenshot 2024-06-11 at 12.11.18 PM.png (172.68 KiB) Viewed 1314 times

These are SfM point clouds and thus using the PM. I selected yes to use the precision maps.

Is there a way to flatten out those spots without losing resolution? Why are they occurring in the first place with no floating points to account for?

Any help with this would be wonderful, thank you!

[daniel]

Sorry, I'm confused: what do you mean by subtracting? Due to how M3C2 works, the 'pinnacle' points are necessarily coming from the original cloud (and from the same cloud as the other points). Isn't it simply noise?

[RiverRunner]

Sorry, subtracting is the wrong term with M3C2, but calculating the difference between two clouds. How do you limit the noise? Is there a step in running M3C2 that I can take? I have tried increasing the diameter of the normals and projection but I get the same thing.

Thanks

[daniel]

So the easiest is to use the 'Tools > Clean > SOR' filter. Or, the scissors tool and do it manually?

[RiverRunner]

Hi Daniel,
I am trying to conduct a volumetric change detection from my two point clouds. I have used the volume tool with CC but I am also interested in using the M3C2 and exporting the raster to analyze in R. When looking at the M3C2-PM I am getting a lot of noise (ranging from -10 to 10m of change) even after the SOR filtering tool BUT I have virtually no noise (-3 to 3m of change) when I run M3C2-PM (vertical). The second range (-3to3m) is more in line with what I observed in the field. If I am doing my analysis on the volume of change is there a difference in what is detected between the default and vertical only M3C2 analysis? Am I losing any possible data by not including the horizontal?
Thanks!

[daniel]

Sorry, I'm not the author of the plugin, and especially not of the 'PM' extension made by James et al. (see https://www.cloudcompare.org/doc/wiki/i ... ision_maps). It may be possible to ask them directly?