3

I have installed Point Cloud Library(PCL) package for using Iterative closest point (icp) my question is: this package could be used for 2D data or not ?I want to align two TSNE data which are 2D.Link of the icp method

user8523104
  • 467
  • 1
  • 3
  • 14
  • 1
    What is the exact error message? Have you installed PCL itself first? – desertnaut Oct 20 '17 at 08:25
  • yes I installed pcl correctly and now I can run this https://github.com/strawlab/python-pcl/blob/master/tests/test_registration.py but the problem is how to use icp successfully, – user8523104 Oct 20 '17 at 10:10
  • so, have you resolved your issue or is the question still open?? – desertnaut Oct 20 '17 at 10:11
  • no I could not solve the question yet I am still trying to call icp method, I was expecting only call icp method but it seems that icp needs Transformation in this link, – user8523104 Oct 20 '17 at 10:13

2 Answers2

2

It is not a problem even if it supported only 3D. You can add a z=0 coordinate to your data points and the z component will always be zero in your solution.

Rooscannon
  • 316
  • 2
  • 7
  • thank you very much for your answer sorry I think icp does work for my data because my TSNE data sets are paired and I found that icp does not work for paired data set. – user8523104 Oct 25 '17 at 17:53
  • You're wellcome. I'm trying to imagine a geometric interpretation of icp of a paired set. I suppose it would just define the x and y translation and a scale factor (rotation), but It wouldn't be best suited to the task if the data plot is not smooth or has gaps so perhaps another approach is better as you said. – Rooscannon Oct 26 '17 at 08:25
  • yes you are right I have applied Procrustes analysis but I am also looking for other algorithm because I am trying to test the stability of tSNE – user8523104 Oct 26 '17 at 16:08
0

PCL does not have a paired data ICP to the best of my knowledge, but it should be pretty trivial to write as the cost function just considers squared distance between each pair.

I would create clouds of the two sets of points, setting z to zero, write your cost function (sum of squared distances of each pair), and then cycle the icp through xy translation steps and z rotation steps using reverse half-ing style icp stepping.

Sneaky Polar Bear
  • 1,611
  • 2
  • 17
  • 29