Automatic transformation matrix

Hi @marcomusy,

I have a couple of point clouds (of a similar object but different type, plains in this case) which are randomly positioned. Now I would like to find the transformation matrix so that all the point clouds to have the same orientation/translation:

from glob import glob
import natsort
import vedo as vd

pcdFiles = natsort.natsorted(glob('**/*.xyz', recursive=True))

for i, pcdFile in enumerate(pcdFiles):
    print(pcdFile)
    s = vd.Points(pcdFile).alpha(0.2)
    vd.show(s, axes=8)

Target position (airplane_0000.xyz):

Transformation to be found:

I am not sure how easy this can be done, possibly using a shape similarity metric or something similar.

testing_samples.zip

[marcomusy]

you can also compute normals for pure pointclouds with myptcloud.computeNormalsWithPCA(). i don’t know how well it works, i’ve never tested it throughly…

[ttsesm]

Thanks Marco, interesting note regarding the hausdorf distance (I was not aware of it, I knew the chamfer distance) I found also an alternative to ICP which takes into account normals as well https://laempy.github.io/pyoints/tutorials/icp.html#NICP-algorithm. This might also get into hand.