LAUSR

Distribution-to-distribution-based lidar odometry is known for its good accuracy, while it cannot run in real-time when the number of points is large. To alleviate this problem, Faster Generalized Iterative Closest Point (FasterGICP) is proposed in this letter, in which an acceptance-rejection sampling-based two-step point filter is proposed to exclude the points that rarely benefit the lidar odometry performance. Specifically, the lidar point cloud is firstly filtered and only the points with high planarity tend to be preserved, which can reduce the distribution approximation errors when the GICP works as a plane-to-plane Iterative Closest Point (ICP). Secondly, during the pose estimation optimization process, the lidar points are further iteratively filtered according to their contributions to the optimization objective function, in which the point's matching error defines the contribution. The two-step filtering process is achieved by designing the target and proposal distributions in the acceptance-rejection sampling framework. With the help of the point filter, our odometry can work in a scan-to-model strategy while demonstrating both efficiency and accuracy improvements. Extensive validation experiments are conducted on the public and our datasets. The results demonstrate that our method can achieve competitive performance compared with the state-of-the-art lidar odometry and Simultaneously Localization and Mapping (SLAM) methods. Our code has been made public available at https://github.com/SLAMWang/fasterGICP.