Efficient LiDAR Odometry for Autonomous Driving

LiDAR odometry plays an important role in self-localization and mapping for autonomous navigation, which is usually treated as a scan registration problem. Although having achieved promising performance on the KITTI benchmark, conventional tree-based neighbor search still has difficulty dealing with large-scale point cloud efficiently. The recent spherical range image-based method enjoys merits of fast nearest by mapping. However, it not very effective to deal ground points nearly parallel beams. To address these issues, we propose novel efficient approach taking advantage both non-ground images bird's-eye-view maps points. Moreover, adaptive introduced robustly estimate local surface normal. Additionally, memory-efficient model update scheme proposed fuse their corresponding normals at different time-stamps. We have conducted extensive experiments benchmark UrbanLoco dataset, whose results demonstrate that our effective.