CloudCompare forum

Hi,

I´m using CC for my master thesis and first I want to say thanks for the wonderful software!

In my work I´m extracting rockfalls by comparing a point cloud (derived by Structure from Motion method) from the year 2000 to recent LiDAR data.
To do the C2M comparison I compute normals for the (quite noisy) SfM point cloud using the height function, radius=0,5m and a prefered orientation to + (0,0,0) which is my scan position. The normals look well oriented for the whole dataset. Then I build the mesh and keep the normals I computed before. Also the mesh seems well oriented.
When I compute the C2M distances from the SfM mesh to the LiDAR point cloud I have my normals showing to opposite directions, even if they are next to each other on the same flat surface.

I tried out to orient the normals with the minimum spanning tree (10 up to 150 neighbors), but the C2M comparison still doesn´t work properly.

I´m sure I just forget one little detail, but I can´t find it out. Anyone has an idea?

Thanks a lot!

What algorithm do you use to compute the mesh?

If the mesh is noisy (it happens a lot with Delaunay as all points are used, even the noisy ones) then the spikes can cause strange/negative values to appear when computing the distances.

If the noise is the issue, then:
- you can try to clean the cloud first
- or use the qPoissonRecon plugin to get a smoother mesh

(in both case you should use the 2.6.2 beta version of CC).

And it the noise is not the issue, can you send me your data?

To compute the mesh I used the Delaunay best fitting plane and further I tried out to clean the point cloud in advance, which was not so successful...
But I think the noise in the cloud really was the problem. Now I used the Poisson Reconstruction and it worked out well. All the normals are properly oriented and I didn´t loose a lot of detail in my point cloud. And I also like the possibility to filter the mesh according to its density values!

Thank you a lot for the help!