Complementing Vehicle Trajectories Using Two Camera Viewpoints

Traffic volume surveying is a crucial activity to get traffic statistics for road management and traffic congestion control. In recent years, the target environment of traffic volume surveying has become more complex, such as the fully automated surveillance of many-way intersections. Further compounding this complexity, some local governments may not be able to install a camera at a high enough elevation to capture the entire intersection due to environmental, legal, safety, or cost restrictions. Therefore, bigger objects such as buses and trucks often occlude other vehicles in the captured image. This occlusion degrades the accuracy of counting and is one of the main problems that makes the automation of traffic counting at intersections difficult. In this work, we propose a Bird’s Eye View (BEV) transformation method capable of 1) removing camera distortion created by wide-angle cameras installed at lower elevations (a common scenario in traffic volume surveys), and 2) utilizing multiple viewpoints to complement object trajectories to reduce accuracy loss caused by occlusion. Furthermore, we evaluate the effectiveness of the proposed method using real-world traffic survey data collected at two intersections in Japan. We find that the proposed method produces a 3% improvement in accuracy over automated counting using a single viewpoint.