Point cloud triangulation with strict boundary

[PhilNamesnik8463]

Hello,

I want to triangulate a simple circle, based on a 3D point cloud, using the MeshLib Python API. However, the triangulation always produces unsatisfying results with lots of holes and the boundary points are not preserved. Is there a way to prescribe certain points for the triangulation?

Detailed problem description:
I sampled 3D points on a circular surface and on the boundary separately and want to conserve the point affiliation as strict as possible. The sampled points look like that:

In this, very simple case, all points lie in the same plane, but this must generally not be the case. I played around with many combinations of triangulationParameters but nothing seems to work, and I always get the following result more or less:

Is there a way to prescribe boundary points if I already know those points before?

Really appreciating your help! Thanks in advance.

[Grantim]

Hello!

Could you please provide pointcloud so we have a look at it?

(Looking at result it seems to me that issue might be related to duplicated points, if so I suggest calling mm.pointUniformSampling or mm.pointGridSampling)

[PhilNamesnik8463]

Thanks for the very quick response. From my workflow it is normally not possible that there are duplicate points, but I uploaded the points as numpy arrays in the attached files.
bd_pts.csv
int_pts.csv

For my workflow it is furthermore necessary to keep track of boundary points and internal points separately which is why I didn't choose the meshlib sampling functions but implemented my own functions. Some other package also "changes" which points are valid and which must be deleted and after this step I have to retriangulate the surface again, in order to know the connectivity/topology of the points.

[Grantim]

I see, thanks for the files.

Here are some suggestions:

1. If you points will always be "terrain-like" on one of the axes: you can use mm.terrainTriangulation function (but note that it suggests Z as depth axis)
2. int_pts - are well sampled and triangulates fine with default parameters, so you can triangulate it separately, then add bd_pts as edge loop in the mesh (mesh.addSeparateEdgeLoop) and stitch it with mesh boundary (mm.stitchHoles). But this requires bd_pts to form a closed loop
3. Changing TriangulationParameters.numNeighbours to 40 seems to work fine in your case

Main reason of the issues with triangulation is non-uniform points distribution, on your case near-boundary vertices should have more neighbors than 16.

Hope it helps!

[PhilNamesnik8463]

Thank you very much! I didn't try that high number of neighbors in triangulation parameters but that solved it. I usually need a very highly resolved boundary which is why the point density is higher in that areas.