Traffic monitoring research based on roadside LiDAR-camera target-free wide-area collaborative calibration

Accurate LiDAR-camera extrinsic calibration is fundamental to roadside multi-sensor fusion for traffic monitoring. Traditional calibration methods usually depend on calibration targets and manual operation, which limits their applicability in large-scale roadside deployments. To overcome this limitation, this paper proposes a target-free wide-area calibration method for roadside LiDAR-camera systems. Instead of relying on artificial markers, the proposed method estimates extrinsic parameters directly from natural traffic scenes. It first extracts complementary geometric and structural features from point clouds and images, and then establishes cross-modal correspondences through a confidence-guided coarse-to-fine matching strategy. To improve calibration stability in dynamic roadside environments, temporal consistency is further introduced into the optimization process together with reprojection constraints. Experiments on a self-built roadside dataset demonstrate that the proposed method achieves an average rotation error of 0.185° and an average translation error of 2.36 cm. Compared with representative target-free methods, it provides higher calibration accuracy while preserving practical computational efficiency. The method also shows good robustness under challenging conditions such as low illumination, occlusion, and dense traffic flow, indicating its potential for real-world roadside deployment.