Problems encountered with ICP algorithm

These days in the study of data correlation method, originally wanted to use the ICP algorithm for distance ranging data registration, but the process of using the problem, the mismatch is not allowed, and the deviation is greater.

The red and yellow 2-D laser points are ICP-registration, but the transformation matrix and yellow multiply after multiplying. Suspicion is because the two point sets only partial data overlap, and the ICP algorithm is best to point can be one-to-ones.

The same problem occurs after using the PCL for point set matching testing.

So I constructed a data, the A-point set of a rigid body transformation, followed by ICP registration, and then contrast, found that the algorithm can be matched well, it should be part of the point set overlap caused.

Problems encountered with ICP algorithm