Hello. I'm new to this forum. I'm searching for other answer on my problem but i didn't find never. So the problem.
I have a mesh and i want to try to remeshing, but my problem is that from wall and road i have a weird mesh. On the images you can see the basic point cloud , normal and the mesh generated. Can you help on find a solution?
P.S. this is a mesh without faces. But i need a solution for real laser scan Point clouds.
Thanks.
Normal and Meshing on oblique
Normal and Meshing on oblique
- Attachments
-
- 1POINTC.jpg (189.58 KiB) Viewed 1826 times
-
- 2Normal.jpg (116.48 KiB) Viewed 1826 times
-
- 4.jpg (254.4 KiB) Viewed 1826 times
Re: Normal and Meshing on oblique
Hi,
What meshing method are you using here? The Delaunay 2D1/2? If yes, then it's definitely not possible to mesh a truly 3D cloud with this method (it's only suitable for almost 2D clouds, such as clouds acquired with an airborne llidar, etc.
For 3D cloud, only the Poisson Reconstruction process is suitable (qPoissonRecon plugin). But for this method to succeed you'll need very clean normals... and the normals you have in the second image seem wrong (all black). As your cloud density is very low you'll need to use a quite big kernel (i.e. radius of the neighborhood sphere - set it so that there's at least 6 points inside). And you'll need to apply the second step to orient all normals correctly (with a minimum spanning tree).
And don't hesitate to test Meshlab, as it is dedicated to meshes (contrarily to CloudCompare which is more dedicated to... clouds ;).
What meshing method are you using here? The Delaunay 2D1/2? If yes, then it's definitely not possible to mesh a truly 3D cloud with this method (it's only suitable for almost 2D clouds, such as clouds acquired with an airborne llidar, etc.
For 3D cloud, only the Poisson Reconstruction process is suitable (qPoissonRecon plugin). But for this method to succeed you'll need very clean normals... and the normals you have in the second image seem wrong (all black). As your cloud density is very low you'll need to use a quite big kernel (i.e. radius of the neighborhood sphere - set it so that there's at least 6 points inside). And you'll need to apply the second step to orient all normals correctly (with a minimum spanning tree).
And don't hesitate to test Meshlab, as it is dedicated to meshes (contrarily to CloudCompare which is more dedicated to... clouds ;).
Daniel, CloudCompare admin