Roadside Bird's Eye View Perception Algorithm for Vehicle Tracking under Vehicle-Road Collaboration

Roadside perception is a crucial research area in the domain of intelligent vehicle-road collaboration. It aims to enhance collaboration between vehicles and the environment perception system while improving perception coverage and accuracy. This paper proposes a two-stage perception and localization algorithm utilizing a bird's eye view from roadside sensing units. The algorithm utilizes an object detection approach rooted in deep learning, which is then followed by a coordinate transformation and correction algorithm. The coordinate transformation is a fusion of the camera imaging principle and inverse perspective mapping. By taking these steps, we can localize human-vehicle objects and construct a car-following model for vehicle-road cooperation using a leader-follower approach. Experimental tests are conducted to verify the effectiveness of the proposed algorithm for perceiving targets and localizing unmanned vehicles following human movement. The results demonstrate that the lightweight bird's eye perception algorithm implemented through roadside sensing units can provide essential perception and location for autonomous vehicle car-following behaviors. This algorithm holds significant potential applications in intelligent parking, obstacle avoidance for car-following vehicles, and inspection of unmanned vehicles.