Fast three-dimensional point cloud registration algorithm based on plane and curvature parameters

The rapid generation of three-dimensional (3D) imaging has improved the safety and operational efficiency of advanced driver assistance systems and mobile robotics technologies. To expedite the stitching process for 3D point cloud data, this study proposes and validates a method that combines coarse registration based on planarity and fine registration based on curvature features. Multiple 3D scenes were collected in the experiments, and the runtime performance was systematically analyzed. Compared with the ICP algorithm, the improved algorithm achieves faster registration speed, with improvements of 19.7% in the cabinet scene, 86.6% in the pig zodiac doll scene, and 61.9% in the tellurion scene. Moreover, the increased efficiency is achieved with negligible impact on the matching residuals. Processing point cloud data without planar information posed challenges in terms of efficiency; however, a notable improvement in processing speed was observed, with point cloud data containing planar scenes. It is important for applications related to 3D imaging registration.