Forum

Author Topic: LIDAR Point Cloud Mesh  (Read 2840 times)

Darko

  • Full Member
  • ***
  • Posts: 128
    • View Profile
    • CADCOM
LIDAR Point Cloud Mesh
« on: September 01, 2020, 11:17:57 AM »
Hi,

As I see there is some problem with the Mesh created from the Lidar Point Cloud. I have tried to make it from the original photo dense point cloud with very nice results. Since I had a dense Lidar scan I assumed that this should give better result, but it didn't.

Check my results on the attached pictures, please. Both meshes were done using the same setup.

Does anyone knows where is the problem and how can it be corrected?

Best, Darko

Arie

  • Full Member
  • ***
  • Posts: 126
    • View Profile
Re: LIDAR Point Cloud Mesh
« Reply #1 on: September 01, 2020, 12:37:49 PM »
Hi,
to be honest, the quality of the LIDAR looks fairly bad, actually even worse than the Agisofts dense cloud. So you can't expect that "bad quality + mediocre quality = good quality".
AFAIK, Agisoft uses poisson surface reconstruction for calculating meshes from points. This algorithm requires good normal estimation to achieve nice results. In this case, it seems the normals were not estimated very well, which led to the "bubble" in the front side.
What's the source of the LIDAR point cloud? You could try using Cloudcompare to estimate the normals (it has different settings, so maybe you will be able to achieve better results).
Cheers.

Darko

  • Full Member
  • ***
  • Posts: 128
    • View Profile
    • CADCOM
Re: LIDAR Point Cloud Mesh
« Reply #2 on: September 01, 2020, 07:31:49 PM »
Thank's Arie,
That was helpful. Now I know where is the problem and I just have to find the best solution for the normals. The lidar scan was done by the flying lidar and the vegetation layer was removed by Global Mapper Lidar option. Seems like I'll have to learn a lot about the Cloud Compare now. Since now we have been using the lidar scan just for the ground surveys and it is the first time we started to play with the 3D models from the lidar point cloud. The idea was to make a solid printable model out of it. Any suggestions?

Best regards,
Darko

Arie

  • Full Member
  • ***
  • Posts: 126
    • View Profile
Re: LIDAR Point Cloud Mesh
« Reply #3 on: September 01, 2020, 07:45:55 PM »
Hi Darko,
IMHO, you will not achieve very satisfying results using that LIDAR scan since the resolution and accuracy seem fairly low. You might have a better chance if you process just the drone images using Metashapes new depth map reconstruction, which achieves more detailed results than using dense pointcloud.
Cheers!

Darko

  • Full Member
  • ***
  • Posts: 128
    • View Profile
    • CADCOM
Re: LIDAR Point Cloud Mesh
« Reply #4 on: September 18, 2020, 11:45:11 AM »
Hi Arie,

The problem is, as you mentioned, the normals and the disperse point clout that cannot be better because of the moving LIDAR nature. The problem that I have with the Photo point clouds is that it doesn't see trough the tree canopy and the walls are not visible. In this case I am missing at least one flight on lower altitude around the object of interest (POI). Anyway, that would not solve the visibility problem.
The Lidar penetrates trough the canopy and wall has some points, not too many, but the plane is defined for the model which I don't know how to make easily. Whatever is suggested as a solution takes a lot of work with software that I am not used to work with.
I am looking for some EASY and fast solution to make a proper STL which can be printed on 3D printer.

Best,
Darko

Arie

  • Full Member
  • ***
  • Posts: 126
    • View Profile
Re: LIDAR Point Cloud Mesh
« Reply #5 on: September 18, 2020, 11:52:14 AM »
Hey Darko,
unfortunately, I don't see an easy and fast approach considering the type of data and your intented use case.
If you don't mind sharing the LIDAR cloud, I'd be happy to take a look at it.
Cheers!

Darko

  • Full Member
  • ***
  • Posts: 128
    • View Profile
    • CADCOM
Re: LIDAR Point Cloud Mesh
« Reply #6 on: September 18, 2020, 02:29:51 PM »
Hi Arie,

Here it is.. https://drive.google.com/file/d/1qDad1Q3TyLoXcseZCLGnFgfKTF0DI9zl/view?usp=sharing
Trees are removed.

Thank's for your interest,
Best,
Darko

Arie

  • Full Member
  • ***
  • Posts: 126
    • View Profile
Re: LIDAR Point Cloud Mesh
« Reply #7 on: September 18, 2020, 07:07:41 PM »
Hey Darko,
attached are some screenshots o the results from some quick processing in cloudcompare. I just calculated the normals with a priority on the z-axis and used the poisson surface reconstruction (afaik Agisoft uses the same algorithm for meshing point clouds).
The mesh would still need to be edited, i.e. removing the lower parts and maybe smoothing the topography out a little. After that you could manually remove the castle from the LIDAR mesh, remove the vegetation from the photogrammetric mesh and merge the two. That would give you a starting point for creating a halfway decent 3D-print.
I would also strongly advise to use the new depth-map based reconstruction for the castle. The resulting mesh should be more detailed.
Cheers!

p.s. Do you want the mesh? Or would you rather recreate it in Cloudcompare yourself?

Darko

  • Full Member
  • ***
  • Posts: 128
    • View Profile
    • CADCOM
Re: LIDAR Point Cloud Mesh
« Reply #8 on: September 27, 2020, 11:39:03 AM »
Thanks Arie,

It's been very helpful. I wasn't aware of the depth maps reconstruction at all. I am reprocessing some older jobs to see what's the difference now. Depth maps are giving nicer results.

Best,
Darko