A Globally Optimal Multi-Constrained Point Cloud Registration Algorithm

Abstract

In response to the issues of lengthy registration times and low accuracy of local optimal solutions, this paper presents a multi-constraint point cloud registration algorithm based on global optimal solutions. Firstly, the algorithm employs the minimum Euclidean distance as the fitness criterion for the Particle Swarm Optimization (PSO) algorithm. A spatial matrix is used to replace the velocity function for iteratively updating the individual best positions and global best position, thereby determining the initial point pairs for point cloud registration. Secondly, the Ant Colony Optimization (ACO) algorithm is adapted by rewriting the pheromone concentration function using the similarity of normal vectors. This step updates the pheromones for the initial point pairs and identifies the point pair set with the maximum pheromone concentration. Spatial transformations are applied to this set to achieve coarse registration. Finally, utilizing curvature similarity error and Euclidean distance as constraint conditions, the algorithm achieves nearest neighbor registration for the point cloud. Experimental results demonstrate that this registration algorithm enhances both registration efficiency and accuracy, outperforming traditional coarse registration algorithms and the original Iterative Closest Point (ICP) algorithm.