Normal consistency based closest point

[Sentient07]

Hello Mike @mikedh ,

I would like to align two meshes that are close in Euclidean space. For this, I'm using trimesh.proximity.closest_point. In my alignment problem, I would like to not just find the closest point, but a point where normals don't differ by 90deg. I would imagine this implies filtering of candidates here (https://github.com/mikedh/trimesh/blob/main/trimesh/proximity.py#L152). However, while doing so, there is a chance that none of candidates satisfy normal constraint. For this, I am assuming that I should expand my search window. Could you tell me how to go about this?

Thanks

[Kiord]

Assuming you are implementing an ICP algorithm, you probably are calling procustes at least once in a loop.

What about setting a low weight when the angle is > 90 deg ?

weights = np.ones(len(candidates))
weights[where angle is > 90deg] = 0.01 # not 0 to prevent breaking when there are less than 3 good pairs
transform = procrustes(source_points, target_points, weights)

I feel like expanding the search window of the closest point algorithm goes a bit against the idea behind ICP. Yet you can try to remove temporarily the "bad" points from one of the meshes and try calling closest_point again until finding enough "good" pairs. It would hurt the performances though.