Real-time Recognition and Reidentification of Vehicles with Matrix Cameras Including Analysis of Simulated Large Camera Distances

Abstract

Traffic data is extracted from video cameras fixed at buildings. The data can be used, e.g., to control traffic lights in future. Cars are reidentified within videos of a second camera monitoring another region. Travel times are obtained which are relevant parameters to optimize traffic control. 3D vehicle models enable extraction of physical dimensions, and vehicle color is obtained for reidentification. Shadows and light reflections on wet streets are corrected. The 3D-concept allows placing cameras in different viewing angles. The algorithms function faster than necessary for real-time applications on a common PC. Results from two cameras show a high reidentification rate. The distance of the cameras is increased within a simplified simulation. The reidentification rate decreases in the mean about 8% when the distance is multiplied by two from about 57m (62.4yd) to about 20km (12.4mi). Mean travel times can be es-timated in different time intervals.