DOF-LIO: LiDAR–Inertial Odometry With Lightweight Dynamic Object Filter

This paper presents a LiDAR-inertial odometry system, DOF-LIO, which enables robots to localise robustly and accurately in dynamic environments. The core of our method is the tight integration of a dynamic object filter within the LIO framework. This filter is evolved from a visibility-based approach, enhanced with capabilities for suppressing false detections and facilitating recovery through the exclusion of outliers and clustering to improve detection accuracy. Additionally, to enable efficient point cloud clustering, we propose a voxel-based clustering method. Both public datasets and real-world experiments validate that our system can effectively detect and filter out dynamic object points, thereby significantly enhancing LIO system accuracy and robustness.