Three-dimensional point cloud registration using particle swarm optimization

Abstract

In order to reconstruct a three-dimensional model, a registration process in the three-dimensional space is required. Although, there are many existing methods in finding transformation matrices, there are none that can handle translation, scaling, rotation, and sheering. Hence, in this paper, we develop a three-dimensional registration method using the particle swarm optimization (PSO). We generate three synthetic shapes, i.e., cube, cylinder, and pyramid. We also transform these shapes to provide the transformation ground truth. The results show that the proposed algorithm is able to find a very good transformation matrix with extremely small registration mean squared errors in the order of 10-31, 10-31, and 10-11 for cubical, cylindrical, and pyramidal shapes, respectively.