Online Joint Calibration Failure Diagnosis of Perception Sensors through Multi-Channel Data Cross-Matching

Joint calibration failures in perception sensors pose a risk to the accuracy of perception outcomes, which can impact safety. This paper proposes real-time sensor joint calibration failure diagnosis models based on multi-channel data cross-matching, including the model based on trajectory matching and the model based on feature matching. The trajectory matching model aligns the output trajectory from the inertial navigation system with the SLAM trajectory generated by the LiDAR, enabling precise matching and calculation of trajectory offsets and rotations. The feature-matching model determines feature offsets and rotations by analyzing the correspondence between shared features captured by the camera and those defined in the high-definition map. Subsequently, this study validates the efficacy of the real-time diagnosis model for sensor joint calibration failures using real-world vehicular data. The experimental findings affirm that the multi-channel data cross-matching-based diagnosis model adeptly identifies sensor joint calibration failures.