Matching Two Spatial Point Clouds Using ICP Algorithm

When processing the registration of two spatial point clouds, we can employ the classic Iterative Closest Point (ICP) algorithm. This algorithm iteratively adjusts the position and orientation between matching points through multiple optimization cycles, minimizing their mean squared distance. The core implementation typically involves three main steps: correspondence estimation (finding closest point pairs using k-d trees), transformation computation (solving for optimal rotation matrix via SVD decomposition), and error minimization. While widely adopted in practice, this approach has certain limitations: it may converge to local minima depending on initial alignment, and demonstrates sensitivity to noise and incomplete data structures. Consequently, numerous enhanced variants have been developed in recent years—such as point-to-plane ICP and robust kernel-based methods—to improve both matching accuracy and algorithmic robustness against outliers and partial overlaps.