Roadside-Onboard Point Cloud Registration for Vehicle-Infrastructure Cooperation Perception in Traffic Collision Zones

Abstract

Vehicle-to-infrastructure (V2I) cooperation perception is considered a promising approach to enhance the perception capabilities of connected autonomous vehicles (CAVs) for achieving high-level autonomy. Point cloud registration serves as the fundamental task in light detection and range (LiDAR)-based cooperation perception. In this study, a roadside-onboard point cloud registration method in traffic collision zones was proposed leveraging the position points of mobile vehicles. First, roadside-onboard LiDAR coordinate systems were aligned using mathematical transformation matrixes. Then, vehicle position points were extracted to fit the centerlines of the lane to form the lane junctions, which were used as reference points in the point cloud rough registration. Finally, the prior feature-based weighted iterative closest point algorithm (PFW-ICP) was presented to achieve a global optimal in point cloud fine registration. To evaluate the effectiveness of the proposed method, the DAIR-V2X dataset and field data were tested in the experiments. The experimental results showed that the proposed method has higher accuracy and robustness compared to other algorithms. The average relative translation error (RTE) was less than 0.55 m, and the relative rotation error (RRE) was less than 0.02° when the ego vehicle going straight, ranging from 0.10° to 0.15° during vehicle turning.