Point cloud registration based on MCMC-SA ICP algorithm

Point cloud registration is very important for workpiece positioning and error evaluation. Generally, the Iterative Closest Points (ICP) algorithm is always adopted as the first choice in fine registration, but requires a more appropriate initial condition to avoid falling into the local minimum. As a result, the ICP shows poor robustness and adaptability when dealing with the point cloud with measuring noise and outliers, complex surface, the far distance between the measured and the design. In this research, an improved ICP method is developed based on the simulated annealing (SA) and the Markov chain Monte Carlo (MCMC) to achieve the global minimum under given any initial conditions. The SA based on Pincus theorem is applied to calculate the rotation angles and translation vectors of the correspondences. To improve the convenience of SA, the MCMC method is applied to improve its sampling ability to satisfy the Pincus distribution. Furthermore, the MCMC-SA ICP algorithm for point cloud registration was designed. Three different kinds of complex curved surfaces were employed to verify the visibility of the developed MCMC-SA ICP approach. By comparing with the traditional ICP, it is indicated that the improved ICP based on MCMC-SA showed better global optimization ability.