HYBRID METAHEURISTIC AND ITERATIVE APPROACH FOR AFFINE POINT CLOUD ALIGNMENT

Abstract

A three-dimensional alignment procedure is necessary to reconstruct a 3-D model. Although there are many methods for determining the rigid and affine transformation matrices of point sets, we are doing a seven-parameter affine transformation that includes scaling, translation, and rotation in this research. Therefore, we hybridize a metaheuristic Particle Swarm Optimization (PSO) algorithm for affine registration with the Iterative Closest Point (ICP) technique for further fine-tuning Point sets to construct a three-dimensional registration approach. Before applying the above algorithms, the point clouds are down-sampled by the voxelization process to reduce their density for a precise transformation. In this research, we have taken two datasets, i.e, Stanford Bunny and Stanford Tyra. To supply the transformation, we additionally alter these shapes. According to the results, with incredibly minimal registration mean squared errors(MSE) of 10⁻²⁶ and 10⁻²⁷. The suggested approach may identify a very good transformation matrix for Bunny and Tyra's point clouds, respectively.