Image pre-registration and multi-criteria mismatch suppression for 3D point cloud registration

In response to the repetitive surface features or symmetrical structures of most parts in the industrial field, with smooth and texture-deficient surfaces, traditional point cloud registration algorithms struggle to extract correct three-dimensional correspondences, leading to misregistration and getting stuck in local optima. This paper proposes a point cloud registration method that utilizes texture image pre-registration and multi-criteria mismatch suppression. Firstly, speckle images are projected onto the object under test, obtaining point cloud and image data from different perspectives.Then, the optimal registration region is located in the grayscale image, and feature points are detected and registered in this region, obtaining corresponding three-dimensional feature points based on the coordinate transformation relationship. Simultaneously, a multi-criteria mismatch suppression method is applied to eliminate mismatched point pairs. Through Singular Value Decomposition (SVD), achieving coarse point cloud registration. Finally, a pose optimization is performed using the Iterative Closest Point (ICP) algorithm on the pre-extracted overlapping regions. Experimental results demonstrate that this method provides efficient and accurate registration for point clouds of parts with surfaces lacking texture and featuring repeated patterns.