Hybrid Global Optimization Methods and Iterative Closest Point on Point-based Approach for 3D Registration

This paper proposes a novel approach to solve pair-wise registration problem which aligns different pointclouds taken from the same object or scenario at different angles. The new method uses points as a searching medium replacing convention six-dimensional one. By doing this, the number of searching dimensions is significantly reduced. Using the same number of searching loops, the new method results in a higher convergence rate into global optimal results. The approach is successfully implemented in a hybrid registration strategy which combines Iterative Closest Point (ICP) as a local aligning tool and a global searching algorithm such as state-of-the-arts including Simulated Annealing, Particle Swarm Optimization, Differential Evolution or a recently developed adaptive Differential Evolution algorithm, ISADE. The accuracy and robustness of the new method over the conventional approach are proved through various experiments on different datasets.