A Multi-View Point Cloud Registration Method for Human Foot

To address the problem of high registration errors in multi-view foot point cloud alignment caused by viewpoint occlusions and other factors, this paper proposes a point cloud registration method that combines the 4-Points Congruent Sets (4PCS) algorithm with the point-to-plane Iterative Closest Point (ICP) algorithm. First, the raw point cloud data collected from multiple viewpoints are preprocessed to remove noise and retain high-quality points. Then, the 4PCS algorithm is applied to perform coarse registration and estimate the initial pose. Subsequently, the point-to-plane ICP algorithm is used for fine registration to obtain the optimal transformation matrix. Finally, the transformation matrices from pairwise registration between consecutive viewpoints (in a clockwise sequence) are imported into Open3D to construct a pose graph, and global optimization is performed to achieve accurate multi-view point cloud fusion. Experimental results show that for three representative viewpoint pairs (B-C, C-D, and D-E), the proposed method achieves RMSE values of 5.706 mm, 5.059 mm, and 6.928 mm, respectively, with registration times of only 1.463 s, 5.019 s, and 3.968 s. In terms of both accuracy and efficiency, the proposed method outperforms conventional methods such as RANSAC + ICP and SAC-IA + ICP, demonstrating its effectiveness and robustness.