I have 2 3D point cloud data sets in cartesian coordinates. They both represent a similar shaped object but are in different positions (one is translated and rotated with respect to the other). I need to write a program to effectively register one to the other. In order to do this, I need an algorithm to determine the "closeness" one is to the other... The points are not paired and since I am dealing with millions of points, it is not possible to measure the distance of each point to all the others. Any thoughts or direction will help!

Question

I have 2 3D point cloud data sets in cartesian coordinates.  They both represent a similar shaped object but are in different positions (one is translated and rotated with respect to the other).  I need to write a program to effectively register one to the other.  In order to do this, I need an algorithm to determine the "closeness" one is to the other... The points are not paired and since I am dealing with millions of points, it is not possible to measure the distance of each point to all the others. Any thoughts or direction will help!

phi · Answer

my first thought is find the center of each cloud  and move both clouds to the origin.

next,  maybe find the principle axis of both clouds ?

bme89 · Answer

That where I was planning on starting from.  I would really like the outer surface of the clouds to match as closely as possible once the program finishes

bme89 · Answer

I have written some code for a centroid calculation and a 3D linear regression to start from

phi · Answer

I assume you have done a literature search ? see https://en.wikipedia.org/wiki/Point_cloud which suggests that a lot of work has been done on this problem.

bme89 · Answer

I found this paper, http://proceedings.spiedigitallibrary.org/proceeding.aspx?articleid=981454 Which is essentially what I want to do, but some of the math gets over my head.