PLACE-LIO: Plane-Centric LiDAR-Inertial Odometry

Abstract

Planes provide effective and reliable constraints for a LiDAR (-Inertial) Odometry method to achieve accurate pose estimation. Typically, one can readily construct local planes by nearest neighbor search or voxelization. Compared to global planes (GPs), these local planes are of lower confidence and always introduce many redundant constraints that may impair the real-time capability. Hence, in this letter, we explicitly extract GPs using a modified uncertainty-guided plane segmentation approach. On this basis, we propose the plane-centric lidar-inertial odometry (PLACE-LIO) method combined with a plane-occupied voxel grid for map representation. Moreover, the proposed LIO system does not solely rely on GPs, which leads to limited applications. We make full use of the scans via a hierarchical data association scheme, and three types of correspondences (i.e., point-to-point, point-to-plane and plane-to-plane) are utilized. We validate the proposed PLACE-LIO on diverse public datasets, and make comparison with other state-of-the-art methods.