Point cloud meshing: Food for thought

[Mesrop (Mesrop Andriasyan)]

I have been looking for available solutions for point cloud meshing in Grasshopper.
So far I could find only the standard Delaunay, Ball Pivot (Milkbox) and Marching Cubes (Cocoon) algorithms implemented in GH. However, there have been some amazing algorithms developed in the recent decade.
Such as those:

• AN EFFECTIVE QUAD-DOMINANT MESHING METHOD FOR UNORGANIZED POINT CLOUDS

• Direct quad-dominant meshing of point cloud via global parameterization

• Surface reconstruction by parallel and unified particle-based resampling from point clouds

• Surface reconstruction from unoriented point clouds by a new triangle selection strategy

• Simple and Scalable Surface Reconstruction

It would be nice to see some of those also developed in Grasshopper. This would be of huge help.

Any volunteers?

[ivelin.peychev]

This would be of huge help.

And a huge investment

[qhongtran]

I have a few point cloud datasets I have done personally. I would be more than happy to share!

[Mesrop (Mesrop Andriasyan)]

Did you do any surface reconstruction? How?

[RIL (Rolf)]

surface reconstruction

MeshLab have a meshlabserver.exe which is very capable, if only some bugs are fixed.

It takes some C++ knowledge to fix, and then it can be run from commandline (and from a component in Grasshopper). I actually to have such a component, but I haven’t been able to fix the bug in the meshlabserver sources.

cnr-isti-vclab/meshlab

Some C++ expert here perhaps can fix the bug(s)? Unfortunately the original developers doesn’t seem to have the time to fix it.

// Rolf

[ahmedshingaly]

i am able to convert point cloud to lines using proximity 3D. the next step would be creating a mesh from lines yet I am unable to do so. do you guys have any idea ar least to convert point cloud to open surface or closed mesh?

[Petras_Vestartas]

If you have Cloud Compare installed you can run it using command line interface (A - cloud, B - normals, C - Mesh).

I was doing this within Rhino using forks (Figure C). Poisson reconstruction does not require any additional software, but it requires normals. If someone knows how to compute normals in the arbitrary pointcloud without using CloudCompare I could share the rest of this method.

For structured clouds it is easy, that comes from Lidar like Faro, but for unstructured normals calculation is hard.

Poisson method outputs distance field, so you can remove mesh faces to obtain mesh for non closed objects.

[valeriacroce93 (Valérie)]

How did you manage to reconstruct the figures in C and D? Did you use Rhino without Grasshopper?

[Petras_Vestartas]

C is poisson reconstruction I have a command line interface for the .exe file. The D is cgal skeletonization using Capybara but it will work on each 5th computer, so I dropped the full workflow since clouds must be perfect, cylindrical fitting could work as well based on 2d approximation 2D skeletonization