Fine Registration Optimization Method for Low-Consistency Point Clouds

Abstract

With the development of three-dimensional point cloud technology, point cloud registration plays a crucial role in computer vision, robotics, and other fields. However, when using sensors to scan the same object, different sets of point cloud data representing the same geometric entity often exhibit low consistency, meaning that there are no direct correspondences between points in the two frames of point clouds. Addressing the precise registration of such low-consistency point clouds remains a challenging problem. Therefore, this paper proposes a refined registration optimization method tailored to low-consistency point clouds. This method compensates for the distance between the points to be registered in the source point cloud and the nearest- neighbor plane in the target point cloud using a curvature weighting factor. Subsequently, a new corresponding point set is constructed for singular value decomposition (SVD), thereby achieving precise point cloud registration. Compared to existing point cloud processing algorithms, this method achieves higher registration accuracy for low-consistency point clouds and demonstrates better applicability.