TriVoC: Efficient Voting-Based Consensus Maximization for Robust Point Cloud Registration With Extreme Outlier Ratios

Correspondence-based point cloud registration is a cornerstone in robotics perception and computer vision, which seeks to estimate the best rigid transformation aligning two clouds from putative correspondences. However, due limited robustness of 3D keypoint matching approaches, outliers, probably large numbers, are prone exist among correspondences, makes robust methods imperative. Unfortunately, existing have their own limitations (e.g. high computational cost or robustness) when facing extreme outlier ratios, unsuitable for practical use. In this letter, we present novel, fast, deterministic guaranteed solver, named TriVoC (Triple-layered Voting with Consensus maximization), problem. We decompose selecting minimal 3-point sets into 3 consecutive layers, each layer design an efficient voting correspondence sorting framework on basis pairwise equal-length constraint. manner, can be selected independently reduced according sorted sequence, significantly lower meanwhile provide strong guarantee achieve largest consensus set (as final inlier set) as long probabilistic termination condition fulfilled. Varied experiments show that our solver against up 99% highly accurate, time-efficient even also real-world applications, showing performance superior other state-of-the-art competitors.