A vehicle tracking system using image processing is being developed which tracks vehicles in video images monitored through a video camera fixed on the road and detects traffic flow and accidents. One of the problems in realizing this system is the difficulty in extracting vehicles accurately using image subtraction between the background image and the input image, when there can be vehicle shadows and illuminated areas. To solve this problem, a method has been developed for separation between vehicle areas and other areas, including vehicle shadows and illuminated areas, using the discrete cosine transform (DCT). This paper describes the method of extracting vehicles using DCT, and the method of tracking vehicles by template-matching using extracted vehicle areas.
Abstract