Skip to Main Content
This paper presents a novel, deterministic framework to extract the traffic state of an intersection with high reliability and in real-time. The multiple video cameras and inductive loops at the intersection are fused on a common plane which consists of a satellite map. The sensors are registered from a CAD map of the intersection that is aligned on the satellite map. The cameras are calibrated to provide the mapping equations that project the detected vehicle positions onto the coordinate system of the satellite map. We use a night time vehicle detection algorithm to process the camera frames. The inductive loops confirm or reject the vehicle tracks measured by the cameras, and the fusion of camera and loop provides an additional feature : the vehicle length. A Kalman filter linearly tracks the vehicles along the lanes. Over time, this filter reduces the noise present in the measurements. The advantage of this approach is that the detected vehicles and their parameters acquire a very high confidence, which brings almost 100% accuracy of the traffic state. An empirical evaluation is performed on a testbed intersection. We show the improvement of this framework over single sensor frameworks.