ACLC: Automatic Calibration for Nonrepetitive Scanning LiDAR-Camera System Based on Point Cloud Noise Optimization

The recent nonrepetitive scanning (NRS)-light detection and ranging (LiDAR), using rosette or Lissajous curves, has higher spatial scanning efficiency and lower price than traditional mechanical LiDAR with fixed scanning angle. NRS-LiDAR has garnered significant attention in the field, particularly in perception research areas such as detection, tracking, and simultaneously localization and mapping (SLAM). However, the NRS mode introduces new challenges, including high-intensity reflections in overlapping areas, nonuniform distribution of point clouds, and scattering at target edges. This article proposes a target-based automatic calibration for NRS-LiDAR and camera (ACLC) system to address the challenges mentioned above and achieve precise calibration of the external parameters for LiDAR-camera fusion. Initially, we model the noise distribution of the point cloud generated by NRS-LiDAR and mitigate systematic measurement errors by suppressing the noise in the point cloud. Next, we employ automatic detection techniques to obtain a stable and accurate point cloud of a checkerboard pattern. Subsequently, we extract the 3-D corner coordinates of the checkerboard by optimizing the point cloud reflectivity intensity using a nonlinear approach. Finally, by utilizing the perspective-n-point (PnP) algorithm with the optimized 3-D corners from the point cloud and the 2-D corners from the camera image, we can determine the external parameters of the LiDAR-camera system. The proposed algorithm's effectiveness and accuracy are validated through qualitative and quantitative analyses of various LiDAR and camera combinations in real-world environments. Additionally, its open-source implementation on GitHub has gained significant recognition within the industry and academia: https://github.com/HViktorTsoi/ACSC.git.