How to evaluate the quality of the 3D model (point cloud)?

[Smejlo]

Hello everyone

I need to evaluate the quality of the 3D model (or point cloud) reconstructed from 3D laser scans. The main issue is, that I don´t have the ground truth model for this evaluation. Can you please suggest any ideas? I read many research papers on which usually used photogrammetry or model created from photographs. But I don´t have photos, I can use only scans from a laser scanner. The main goal is to find the best reconstruction method (I have to try many Iterative Closest Point (ICP) variants).

I am thinking about compare distances between reference points of the reconstructed model with manually measured distances from the real object, but it should be an automated process. Or maybe I can use more scanners in fixed positions for scanning different objects and then evaluate ICP transformation. The most consistent transformation should be the best. What do you think about it? Please write your ideas. Thanks

[Micr0]

Hello everyone

I need to evaluate the quality of the 3D model (or point cloud) reconstructed from 3D laser scans. The main issue is, that I don´t have the ground truth model for this evaluation. Can you please suggest any ideas? I read many research papers on which usually used photogrammetry or model created from photographs. But I don´t have photos, I can use only scans from a laser scanner. The main goal is to find the best reconstruction method (I have to try many Iterative Closest Point (ICP) variants).

I am thinking about compare distances between reference points of the reconstructed model with manually measured distances from the real object, but it should be an automated process. Or maybe I can use more scanners in fixed positions for scanning different objects and then evaluate ICP transformation. The most consistent transformation should be the best. What do you think about it? Please write your ideas. Thanks

If I understand you correctly, the easiest would seem to be to compare measurements between know points. I often measure off machined surfaces with the CMM or even calipers. It can be fairly easy to get an understanding or the 3D accuracy of your scans that way.

[Smejlo]

Thank you for your response.

I will try to compare measurements between known points on the model in real-world and model reconstructed from scans. If you have other ideas (the best solution for me is to create an automated process for quality evaluation), share them, please. Thanks

[mading]

GOM Inspect. Free. Tou can compare nominal geometries with actual scans.

[Smejlo]

Thank for your response.

It looks like useful software for my work. If you have other tips, please share them. Thanks