Reconstruction of non-textured objects

[promistrio]

Hi! I tried to reconstruct a mobile tower using a Meshroom. And I found that this tower was completely ignored during the reconstruction of the depth map. Example of source images:

Sorry, but now you can see the key features of the image. I will post a depth map later, but the tower is black on the depth map. For example, Metashape can reconstruct it, but sometimes with a different quality. or

The Meshroom ignores this. Can it be customized? I think it can be reconstructed from video using LSD-SLAM but it completely another way. I'm trying to create a way to do this. Maybe use the key points of geodesk or implement other improvements(like https://github.com/GhiXu/ACMM), but I can't yet suggest the optimal way.

[fabiencastan]

Hi, That's a very challenging object to reconstruct due to the large textureless areas! The usual recommandation for textureless areas is to project a pattern, obviously not possible outdoor here. Or to shoot much closer to get surface details, but the size of the object makes it impossible here. And it's both a problem of SfM and MVS. As the poses of the cameras will not be precise enough if they rely on feature points extracted only on the ground.

What version of Meshroom are you using? It would be interesting to see if the "develop" branch does not improve the situation in comparison to the latest release with the various changes on features extraction.

And I would still recommend to shoot a bit closer. Could you also shoot with the camera in a vertical orientation to have more useful pixels in each image? Best,

[promistrio]

Hi, That's a very challenging object to reconstruct due to the large textureless areas! The usual recommandation for textureless areas is to project a pattern, obviously not possible outdoor here. Or to shoot much closer to get surface details, but the size of the object makes it impossible here. And it's both a problem of SfM and MVS. As the poses of the cameras will not be precise enough if they rely on feature points extracted only on the ground.

What version of Meshroom are you using? It would be interesting to see if the "develop" branch does not improve the situation in comparison to the latest release with the various changes on features extraction.

And I would still recommend to shoot a bit closer. Could you also shoot with the camera in a vertical orientation to have more useful pixels in each image? Best,

I have more than 600 photos on the object. At an angle of 60 degrees and perpendicular. The object also has antennas, but the pole is the worst everywhere (metashape, meshroom) . We want to start using 3D TOF cameras in the near future. Do I need to build a new branch or are there links to the executable versions?

And I would still recommend to shoot a bit closer.

I tried. Depends on the quality of the tower and the background on the ground. It is strange that mushroom does not use such objects, and does not try to add them to depth maps.

[promistrio]

I used meshroom version 2021.1.0. sample of depth maps:

[promistrio]

And I would still recommend to shoot a bit closer.

I think doesn't help me, may be in good condition. You can see and other. Sometimes, of course, there are interesting objects, but this happens rarely

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.