The need to calibrate the camera is present in most computer vision traffic monitoring systems. This, in principle, simple process gets complicated by a number of undesirable effects incurred by the perspective of the camera. In this paper we analyse the circumstances surrounding the camera calibration process in a typical traffic monitoring application and present an integral method that will obtain the transformation matrices required to calibrate the camera and construct, in real time, a rectified image, where perspective effects on the road plane have been eliminated. The use of this perspective-free image will greatly simplify subsequent tracking stages since vehiclesÆ shapes can be approximated by rectangles and parallel trajectories will be restored. (Author/publisher).
Abstract