An Improved Point Cloud Registration Strategy based on Simulated Annealing and Barzilai-Borwein Gradient

Three-dimensional (3D) point cloud registration is a fundamental problem in computer and robotic vision. The iterative closest point (ICP) is the most widely used registration method but it usually suffers from local minima. To address the local minima of ICP without increasing too much computational cost, we propose an improved ICP algorithm for fast global 3D point cloud registration based on simulated annealing (SA) and Barzilai-Borwein (BB) gradient. In this scenario, simulated annealing is adopted to solve the local minima problem of ICP, and BB gradient to improve the calculation efficiency. Furthermore, during the executive process of the algorithm, we have set flexible termination conditions for the loop to relieve its computational burden. Extensive experiments show that our method ensures competitive performance in precision to the typical similar registration algorithms, and its efficiency is improved evidently.