ICP registration issues

[Samdesi03]

Hi,

I am doing a series of analyses in which I compare points clouds generated through TLS and SfM-MVS. The first two registration with ICP gave no issues (I am doing ICP because I wanted the matrix for the translation and rotation values), but the last two sets of point clouds don't want to work. When I first align them using the point-pair tool all good. When I go and apply the ICP my point cloud actually shifts (about 10 cm).
Do you know how I could get around this issues/why is doing that?

Many thanks.
Sam

[daniel]

Is the overlap parameter set correctly? In case of doubt, try with 50% ;)

If the two clouds don't overlap (more exactly, if there are points in the aligned cloud that don't have equivalents in the reference cloud), then it can shift the alignment because ICP will try to minimize the distances over the whole cloud.

Another potential issue is noisy points that are hanging 'in the air' (you can increase the point size to check that). But the 'overlap' parameter should help with that as well.

And if it's not working better, can you post a snapshot of your clouds?

[Samdesi03]

Hi Daniel,

I think the issue was due to the luck of equivalent points in the reference cloud (the two acquistino strategy captured different areas around the subject). Once I excluded those areas worked fine :)
Thanks so much for the help!

Best,
Samantha