Point Cloud Registration Method Based on Regional Centroid in Conjunction with Principal Component Analysis

In view of the problems of traditional iterative closest point (ICP) registration algorithm being sensitive to initial pose, prone to local optima, and characterized by low computational efficiency, this paper proposes a point cloud registration method based on regional centroids combined with principal component analysis (PCA). Firstly, extracted based on local normal vector variations of the point cloud, feature point sets are divided into three regions using the k-means clustering algorithm, and an exhaustive method establishes spatial correspondence between source and target point cloud regions. Then PCA is applied to adjust the initial pose by using regional centroids. Finally, a k-d tree-accelerated ICP algorithm is introduced for fine registration. Experiments on the Bunny model demonstrate that the proposed method achieves stable registration under small, medium, and large angular deviations. Compared to the optimal baseline method, the computational time is reduced by approximately 15%, validating its comprehensive advantages in robustness, accuracy, and efficiency.