A point cloud registration method combining enhanced particle swarm optimization and iterative closest point method

摘要

The iterative closest point algorithm (ICP algorithm) is commonly applied to spatial registration of three-dimensional (3D) models when a rough and initial estimation of the relative pose is known. Many variants of the ICP algorithm have been invented to affect all stages of the algorithm from selection and matching points in order to meet the minimization strategy. In this paper, a good initial relative position of two sets for the conventional ICP algorithm has been provided by the particle swarm optimization algorithm (PSO algorithm). The combination of the two algorithms is used for avoiding the local optimal if the two sets of point cloud are extremely different in initial directions and positions. Compared with other PSO-ICP algorithm, the advantages of this paper are in two aspects. One is that we constrain not only the distances between the point pairs but also the directions of their normal vectors. Another is that the outlier detection and filtering of scattered point cloud by classification have been applied to balance between accuracy and speed of the registration. The experimental results show the effectiveness of our method, without detecting the geometric features and the overlap of the two sets conveniently.