A robust point cloud registration algorithm based on inlier detection

Registration is the most fundmental and important task when processing point-clouds. In correspondence-based alignment of point-clouds, using point features to match correspondences can result in a very high outlier rate. Current methods still have low efficiency, recognition rates and accuracy. In this paper, we propose a point-cloud descriptor based on SIFT feature points and a method for eliminating outliers based on similar triangles. Firstly, the SIFT algorithm is used to extract the feature point-cloud, and then the FPFH feature descriptor is calculated. After obtaining the initial matching set K, the refinement strategy is used to further filter the outliers and improve the quality of internal correspondence. Finally, using the improved correspondence, the final transformation matrix is calculated. This method can achieve fast and accurate outlier removal and stepwise correction parameter estimation. Experiments on a large number of simulation experiments and real data sets show that with the proposed method, effective point-cloud registration can be achieved even if a high outlier ratio is exceeded.The proposed method enables effective registration even when the outlier ratio is high and the efficiency reaches a high level.