An Extrinsic LiDAR-GNSS/INS Registration Approach for Autonomous Driving

Ensuring precise registration of sensors stands as a pivotal element in integrating LiDAR and inertial data for autonomous driving. This study introduces an innovative three-phase method for extrinsic registration among GNSS/INS and LiDAR systems in the scenarios of autonomous vehicles. The initial phase rapidly refines extrinsic parameters among sensors by utilizing point cloud surface features, swiftly narrowing down the initial large error to a significantly smaller range. Subsequently, the second phase further refines these parameters by considering LiDAR-mapping space occupancy while eliminating any distortion caused by motion. The concluding phase rectifies z-axis errors attributed to the vehicle's plane motion, resulting in the attainment of precise extrinsic parameters. Notably, our method leverages planar environmental features, facilitating swift registration procedures. Evaluation on real-world datasets substantiates the reliability and accuracy of this approach.