Leveraging LiDAR Reflectance Images for Sparse Odometry

Abstract

Accurate odometry is essential for Simultaneous Localization and Mapping (SLAM), yet traditional methods relying on geometric features struggle in feature-poor environments such as tunnels. Our work addresses this issue by leveraging LiDAR reflectance data to develop a robust odometry technique. This approach generates reflectance images, extracts 3D keypoints, and employs an IMU-based outlier detection process eventually refined by RANSAC algorithm. Unlike geometry-based methods, our solution can operate in highly symmetrical environments, producing consistent trajectories even where geometric methods fail. Our evaluations highlight the capacity of our approach to maintain trajectory coherence in GNSS-denied and geometrically degenerate scenarios. This robustness underscores its potential for reliable navigation and mapping when traditional SLAM solutions are inadequate.