inspired by http://stackoverflow.com/questions/20120384/iterative-closest-point-icp-implementation-on-python
Call like this,
ret = icp(d1, d2).
d1, d2 are numpy array of 2d points.
The return value ret is the convert matrix with 2 rows and 3 coloums.
icp estimates rotation, moving, scaling(each x and y Separately) convertion.