Fast Point Ranking - Robust Cloud Voxelization and Denoising for Lidar Odometry and Mapping in Adverse Weather Conditions

Abstract. Lidar Odometry (LO) is crucial for autonomous navigation, forming the foundation for simultaneous localization and mapping, and providing essential feedback for control systems. Adverse weather conditions, however, introduce false readings, missing echoes, and noise to lidar measurements, severely degrading point cloud quality and compromising LO effectiveness. This study proposes Fast Point Ranking (FPR), a technique that effectively minimizes the impact of adverse weather effects during registration and map denoising via a robust rank-based point cloud voxelization. Experiments on the real-world KITTI-360 and the novel, openly shared Adverse-Weather-KITTI-360 dataset demonstrate that FPR significantly enhances localization accuracy in adverse weather, providing up to 10 m smaller root mean square errors in positioning. Furthermore, FPR shows increased resilience to adverse weather, maintaining consistent localization accuracy despite the weather conditions.