Non-Rigid Alignment to a reference cloud

[Andy99]

Hi Everyone,

I am looking for an option for non-rigid alignment of mobile laser scans (GeoSLAM ZEB) does anyone know if that’s possible?

We have scanned the interior and exterior of a long service depot (250m+) with a GeoSlam ZEB scanner and ended up with large errors and it seems that it is impossible to align them correctly.

I also have some overlapping data from a NavVis scanner and a good photogrammetry (exterior only) data from what seems to be correct, they have aligned nicely but they don’t cover everything and I would really need some of the GeoSLAM scans to fill the missing parts.

I wonder if there is a software or there is a way to "warp" the non-accurate GeoSLAM point clouds to a reference cloud coming from other sources on a non-rigid way?

[smacl]

We support a very wide range of transformations for point clouds in SCC including 3d conformal (Helmert) which is possibly what you're after. This will compute a transformation based at least 3 common known stations or points between your two models. If you want to hold your levels fixed and only want rotations in plan, you're better off with a 2d conformal transformation. If you're scale is non-uniform, a 2d affine transformation might be a better option.

If you need to apply different local transformations to different parts of your Geoslam point cloud data, these types of transformations won't work and you need to correct the trajectory and re-register.

[kpob46]

I believe GeoSLAM does have the ability to complete a non-rigid alignment, but it would have to be completed on surveyed targets for the non-rigid transformation to take place during processing.

Alternatively (and painfully), I have sectioned out some bad GeoSLAM data based on time in CloudCompare. I then took the useful data and C2C registered it to other overlapping datasets.