Is it possible to use marching cubes mesh extraction for xyz point cloud?

Question

Is it possible to use marching cubes mesh extraction for xyz point cloud?

Closed this issue 3 years ago · 8 comments

1939938853 commented 4 years ago

Hi,

Instead of RGBD, just wonder it is possible to use marching cubes mesh extraction for xyz point cloud?

Thanks

Answer 1 · 2020-09-22T12:00:06.000Z

Hi @1939938853 ,

Thank you for your feedback.
Do you want to generate a mesh from a point cloud?
Poisson's surface reconstruction seem to be better than marching cubes.

https://towardsdatascience.com/5-step-guide-to-generate-3d-meshes-from-point-clouds-with-python-36bad397d8ba

These are not implemented yet.

Answer 2 · 2020-09-22T14:40:42.000Z

I have tried the open3d Ball-Pivoting and Poisson. Plus normal estimation and normal rearrangement, it is difficult to achieve desired performance and rendering effect.

The Marching Cubes looks promising. But may need to turn a point cloud into a (T)SDF volume?

Answer 3 · 2020-09-22T17:07:03.000Z

From here, it looks that PCL can use Marching Cubes to create mesh?

Answer 4 · 2020-09-23T12:15:27.000Z

Thanks for the reference.
It looks like PCL voxelizes the point cloud and applies Marching Cubes.
I add it to a future feature.

Answer 5 · 2020-09-23T16:28:08.000Z

Marching Cubes based surface reconstruction seems promising in the aspect of real-time related performance.

Can avoid normal estimation? Normal estimation is not so "cheap", also GPU acceleration is not so effective, especially when the point cloud is not so big. Furthermore to make all normal direction towards outside is not so "cheap" either.
From Nvidia's early presentation, it seems that Marching Cubes is parallelization friendly?

Surface reconstruction from point cloud, except meshing, perhaps the other better choice is the surface splatting? I guess that the Cupoch visualization of point cloud with normal has some kind of splatting? But it is not "water-tight". If Cupoch can also realize the surface splatting as shown in the this link , that could be great. Even though surface splatting require per-calculated normal, but seems that it is not sensitive to the normal direction.

Answer 6 · 2020-09-24T23:38:33.000Z

Thank you for the some suggestions.
I didn't know about surface-splatting.
The marching cubes sounds advantageous from the point of view of the speed of the calculation.
I'll look into surface-splatting as well.
Thanks!

Answer 7 · 2020-10-27T07:48:08.000Z

Hi,
Are you planning to add the extraction of MarchingCubes from ScalableTSDFVolume? Since I think we can reconstruct larger area of the environment.

Answer 8 · 2020-10-27T13:13:18.000Z

Yes,
I are planning to add it.
Please wait for a while.