CloudCompare forum

So I had a look at the tutorial on YT for Meshing Point Clouds in Cloud Compare, the final part shows the creation of 2 meshes, one from the original and one from a copy of it. The third mesh generated looks excellent but I'm not able to re-create the same result by removing the bubbles. The cloud starts at 77M and I subsample it down to 22M. This speeds up the process and it seems I was able to get a much better mesh from the 77M but it had HUGE bubbles inside the interior of the point cloud that I could not fix. Especially since I was unable to find a way to navigate the camera into the cloud. I would love some pointers if possible and if you're interested in trying it yourself here's a link to the model.

https://www.dropbox.com/scl/fi/nmmkbke1 ... 50j5l&dl=0

So to navigate the camera inside the cloud, you need to switch to the 'Viewer based perspective' mode, and then use the mouse to enter 'inside' the cloud. You can also play with the 'min clipping distance' (via the camera parameters) to clip the points that are too close to the camera.

And about the bubbles, it's all about how clean/correct your normals are. So I guess the issue is with how your normals are computed. But this can be really challenging depending on your cloud. I'll try to take a look when I can.

Ok, so I looked at the cloud, and it's pretty crowded inside the building!

There's also a single sensor defined in the E57 file, while there are clearly much more scans. This can be very misleading for CC.

So what I did on my side:
- I removed all the spurious points outside of the building (you can either use the scissors tool or the Cross Section tool)
- I computed the normals with the following parameters: (mind the fact that 'Use sensor(s) whenever possible' is disabled - instead I told CC to orient the normals so that they all point towards the center of the room, which is clearly a big approximation but probably more robust than the other heuristics in this case).

This gave decent results: And you get a not so bad result with Poisson (considering the challenge ;): To compute good normals and get a better result, CC will need some additional info. Do you maybe have the individual scans? Or at least a way to export the cloud with its individual scanner positions? Or equivalently the individual scan structures (as grids). Normally formats like E57 or Faro's proprietary format are able to preserve this information. With this, CC would be able to compute clean normals everywhere, and the mesh quality will be improved.