Onboard LiDAR–Camera Online Calibration With Pavement Marking Corner Points for Pavement Distress Fusion Detection

Multi-sensor fusion is considered a cost-effective way to enhance accuracy, robustness, and comprehensiveness for pavement distress detection in complex traffic environments. Sensor calibration is a fundamental task for data synchronization from heterogeneous sensors to ensure precise and reliable fusion. However, to the best of our knowledge, efficient and robust online calibration for pavement distress detection using onboard camera and LiDAR sensors has not been explored. Therefore, a novel LiDAR-camera online calibration framework was proposed by detecting the pavement marking corner as the feature to establish correspondences for data alignment. First, a uniform region of interest (ROI) projection method was presented to align pavement coverage areas detected by LiDAR and camera for pavement distress fusion detection. Then, pavement marking corner detection algorithms were developed to extract pavement marking corners from LiDAR and camera data, respectively. In addition, a pavement marking corner point binary coding algorithm was developed to match the corner points detected by LiDAR and camera for establishing correspondences in external parameter estimation. Finally, an improved differential evolutionary (IDE) algorithm was developed to optimize the LiDAR-camera calibration parameter by integrating a priori knowledge during initialization to expedite the global search process and reduce computational resources. Experimental results showcased the proposed method outperforms the comparison method, achieving the average point-to-point, point-to-line, and line-to-line reprojection errors of 2.59 pixels, 3.98 pixels, and 1.08°, respectively. It demonstrated its accuracy and robustness in different pavement distress conditions and pavement marking clear and distinct conditions.