Roadside camera calibration and its application in length-based vehicle classification

Video surveillance system is widely used in the current traffic monitoring system. Existing commercial image processing system works well in vehicle detection and tracking, but it has difficulties in vehicle classification if the camera is not calibrated. A priori camera calibration is used in many systems, which is a complicated process. In this paper, a very simple roadside camera calibration is proposed. Our algorithm is based on some reasonable assumptions, which only requires a traffic video for calibration. The camera is calibrated by estimating the lane boundaries and the vanishing point. Then the position in image coordinate is transformed to real-world reference systems. After calibration, the vehicles are classified by their lengths on the road. Finally, the performance of our classification approach is illustrated and discussed.