DANIEL: A fast and robust consensus maximization method for point cloud registration with high outlier ratios

Correspondence-based point cloud registration is a cornerstone in computer vision, robotics, autonomous navigation and remote sensing, which seeks to estimate the best rigid transformation between two clouds from correspondences established over 3D keypoints. Owing limited robustness accuracy, current keypoint matching techniques are rather prone yield outliers, probably large numbers, thus making robust estimation (outlier rejection) for problem very important. Unfortunately, previously proposed methods often suffer high computation cost or insufficient when encountering (or even extreme) outlier ratios, hardly ideal enough practical use. In this paper, we present novel, time-efficient consensus maximization solver, named Double-layered sAmpliNg with based on stratIfied Element-wise compatibiLity (DANIEL), registration. DANIEL smartly designed layers of random sampling operation, order find inlier subset lowest computational possible. Specifically, (i) apply rigidity constraint prune raw outliers first layer one-point sampling, (ii) introduce series stratified element-wise compatibility tests conduct rapid checking minimal models so as realize more efficient second two-point (iii) employ probabilistic termination conditions ensure timely return final set. By conducting experiments multiple real datasets, demonstrate that solver against 99% also significantly faster than previous state-of-the-art algorithms, fairly use real-world applications such object localization scan problems.