CloudCompare & PCL ICP registration (point to face)

I. Introduction

The ICP algorithm is divided into 6 stages, as shown in the following figure:

(1) Select the overlapping point cloud subsets. In this step, if the amount of original point cloud data is relatively large, the original point cloud will generally be down-sampled.
(2) Match feature points.Usually it is the two closest points, but of course this depends on the criteria of judgment.
(3) Weighting.The found corresponding points are weighted according to how well the points match.
(4) Suppress matching points.Some poor quality point pairs are suppressed (rejected) according to the matching degree of matching points.
(5) Error minimization.The transformation parameters are estimated by minimizing the sum of squared distances.
(6) Point cloud transformation.Transform the source point cloud by the estimated transformation matrix.

Except for the last step in the whole process, the remaining steps have been explored and researched by a large number of literatures, so there are many variant methods, among which point-to-surface ICP is one of them.

2. Types in PCL