The goal of ICP is to find the rigid transformation T that best aligns a cloud of scene points with a geometric model.
The alignment process work is to minimize the mean squard distance between scene points and their cloest model point.
ICP is efficient, with average case complexity of O(nlogn) for n point images, and it converges monotonically to a local minimum.
At each iteration, the algorithm computes correspondences by finding cloest points, and then minimizes the mean square error in positon between the correspondences.
Since ICP is an iterative descent algorithm, it requires a good initial estimate in order to converge to the global minimum, and all scene points are assumed to have correspondences in the model.
