Global Iterative Closet Point Using Nested Annealing for Initialization

In computer vision, Iterative Closest Point (ICP) has been a key tool for registration algorithms, a fundamental task in computer vision. However, ICP based registration algorithms always face with local minima problem and pre-aligned pointsets are the must to guarantee correct convergence. Pre-alignment used to be carried out by our human in some mesh processing softwares. This paper provides a solution for initialization problem for registering two 3D surfaces under L2 error using ICP algorithm. Our algorithm uses a combination between Nested Annealing (NA) and ICP in which NA is used as global optimization search engine to find the global minima with a novel approach of using point based boundary searching. The algorithm uses ICP to derive local minima as well as local minima error. The integration between ICP and NA is successfully implemented and coded into a program which inputs two range image and outputs the transformation matrix between them at high accuracy and success rate.