Fast Correspondence Clustering by Cluster-Wise Spatial Compatibility for Point Cloud Registration

Spatial compatibility (SC) is a widely used similarity measure in outlier rejection methods. However, the single compatibility check used in the spatial compatibility measure is vulnerable to problems such as feature ambiguity, resulting in inaccurate or even incorrect alignment. To solve this problem, we extend the common spatial compatibility measure to correspondence cluster and propose the cluster-wise spatial compatibility (CSC) measure, which eliminates outliers by performing multiple compatibility checks between correspondence clusters. Then, to better utilize the advantages of CSC measure, we design an innovative and efficient correspondence clustering strategy to group the initial correspondences into different clusters. Finally, we use weighted SVD algorithm to solve the rigid transformation from reliable clusters and select the best model as the result. Sufficient experiments show that our method achieves excellent performance in both outdoor and indoor scenes.