Multi-view Joint Online LiDAR-Camera Extrinsic Calibration

Accurate extrinsic calibration of LiDAR and camera is crucial in multimodal perception systems. However, existing calibration methods focus on the calibration between a single camera and a LiDAR, overlooking the capability of using multi-view perspectives to eliminate defects in single-view images. In this work, we propose a novel automatic LiDAR-camera calibration method for sensor systems equipped with multiple cameras and one LiDAR. First, we propose a grid-level method to establish the correspondences between images and point clouds, allowing the pixel coordinate of the projected point cloud to be continuously differentiable. Then, we apply a multi-view joint semantic and intensity consistency (MJSIC) score to evaluate the quality of the extrinsic parameters. Finally, we propose an adaptive score gradient optimization (SGO) algorithm based on the gradient of the MJSIC score to compute the extrinsic parameters with minimal computational resources. To verify the benefits of our method, we conducted experiments on the KITTI odometry benchmark dataset. The experimental results demonstrate the accuracy and robustness of the proposed approach, as well as its superiority in terms of time efficiency.