Image and Point Cloud Frustum-Based Auxiliary Localization for Agents in Complex Scenes
Yonghui Huang, Xi Zhang, Baixuan Zhao, Qi Jiang, Qingfeng Ou, Chuan Hu · IEEE Transactions on Instrumentation and Measurement · 2025
Accurate positioning is crucial for targets like autonomous guided vehicles (AGVs). While Global Navigation Satellite Systems (GNSS) positioning is commonly used for AGVs outdoors, obscured environments pose challenges to accurate positioning. To address this issue, this paper proposes an auxiliary positioning method for vehicles based on roadside devices. Firstly, the roadside camera detects the target vehicles. Secondly, the point cloud corresponding to the bounding boxes of the target vehicles is obtained based on the roadside lidar. The positions and heading angles of the target vehicles are determined through 3D detection algorithms within the point cloud cones. Next, the vehicles’ future states are predicted based on their historical positions and heading angles. Finally, the information of the target vehicles is transmitted through communication units. This method is implemented and tested on a roadside computing platform, conducting real vehicle experiments. The results demonstrate that the average positioning error is less than 1.0m, affirming the effectiveness of this method. Overcoming the limitations of GNSS in obscured environments, this auxiliary positioning method has significant practical implications for AGVs.