A 3D Registration Algorithm Based on Mesh and Common Subgraph

In the fields of computer vision and 3D modeling, Point cloud registration (PCR) plays a crucial role, involving the alignment of multiple point cloud datasets acquired from different perspectives or time points into a common reference coordinate system. It is vital in various applications such as 3D reconstruction, object recognition, and autonomous driving. However, the process of PCR often encounters challenges such as noise interference, partial overlap, and data loss, which significantly impact the accuracy and efficiency of registration. This paper addresses these challenges in PCR by proposing an algorithm that utilizes graph structures. The algorithm converts point clouds into mesh form, which avoids intersecting edges in space, and then filters and removes certain edges based on whether the curvature difference between two points exceeds a set threshold. This ensures that the remaining edges can effectively represent the local edges of the object, thus obtaining a graph representation of the point cloud. Subsequently, for the graphs corresponding to the two point clouds, an optimized graph algorithm is used to search for common subgraphs to determine the correspondence between points in the point clouds, achieving point cloud registration. Experimental results demonstrate that compared to other algorithms, our approach yields satisfactory performance.