Roadside LiDAR-Camera Fusion Detection Based on Spatiotemporal Calibration

摘要

Multi-sensor fusion can take the strengths of individual sensors to adapt complex environments and address target detection challenges. Therefore, an efficient and accuracy roadside LiDAR-camera fusion detection method was proposed. First, a random sample consensus (RANSAC) and Levenberg-Marquardt (LM) joint algorithm was developed to align LiDAR-camera spatial coordinates before leveraging the temporal synchronization mechanism of robot operating system (ROS) to align the data timestamp. Then, traffic objects were detected using LiDAR and camera sensor data with improve density-based spatial clustering of applications with noise (DBSCAN) algorithm and YOLOv10 framework, respectively. Finally, an improve Hungarian algorithm combined with KD-tree matching strategy was proposed to perform decision-level data fusion. Experimental results showed that the proposed method achieves a position error of 0.32m, a translation error of 0.09m, and a heading angle error of 2.33° in spatial calibration and a mAP of 91.7% in detection and classification, which demonstrated its accuracy and robustness in different traffic scenarios.