Automatic Extrinsics Calibration of LiDAR and Camera Based on Sign Detection in Subway Depot

Accurate LiDAR-camera extrinsic calibration is essential for robust sensor fusion perception in self-driving subway systems. However, traditional target-based methods are impractical for urban rail environments and cannot support online calibration. Environmental feature-based and learning-based methods also face challenges such as low robustness in sparse LiDAR data and poor generalization to new scenarios. To address these challenges, this paper presents a novel automatic LiDAR-camera extrinsic calibration method designed for subway scenarios. Leveraging the consistent presence of specific signs—such as rail number signs, buffer stop signs, and overhead line end signs, we detect corresponding keypoints in both camera images and LiDAR depth-intensity fusion images using a unified YOLO detection model. These matched keypoints are then used to estimate extrinsic parameters. Our method avoids the need for external calibration targets, supports online calibration and correction, and offers improved robustness and accuracy. Experimental results demonstrate the effectiveness of the proposed approach in achieving reliable calibration under real-world subway operation conditions.