Fault Detection and Exclusion for Robust Online Calibration of Vehicle to LiDAR Rotation Parameter

LiDAR (Light Detection and Ranging) is a technology that is widely used in Advanced Driver Assistance Systems (ADAS) and Autonomous Driving (AD) applications for tasks such as perception, localization, and Simultaneous Localization And Mapping (SLAM). For LiDAR to function appropriately, it is necessary to calibrate the rotation extrinsic parameters that define the orientation relationship between the LiDAR and vehicle coordinates. These parameters include roll, pitch, and yaw, initially calibrated during manufacturing. However, they may change over time due to vibration, heat, loading, or accidental impact during long-term vehicle operation. Using parameters that do not reflect these changes can drastically degrade the performance of ADAS and AD applications. This paper proposes a precise online calibration process to detect and correct LiDAR rotation parameter changes. The precise online calibration system utilizes standard on-road driving data to estimate the LiDAR-vehicle rotation extrinsic parameters online. The proposed process consists of two parts: roll-pitch parameter estimation and yaw parameter estimation. The system estimates the roll-pitch relative to the ground plane in the roll-pitch estimation. In estimating the yaw parameter, a simplified hand-eye calibration approach is used, which leverages vehicle and LiDAR odometry to estimate the yaw of the LiDAR calibration parameter. To improve the accuracy and stability in the yaw estimation, a fault data exclusion algorithm is introduced to identify and exclude faulty inputs based on the difference between the vehicle and LiDAR odometry. The proposed system's effectiveness, robustness, and accuracy are verified in various environments, including urban roads and highway scenarios.