Measurement of heavy traffic using temporal template

This paper presents the method which can measure traffic situation in heavy congestion. The proposed method calculates average speed of passing vehicles from Temporal Template well used in the field of gesture recognition. Temporal Template represents a history of motion as a gray-scale image. It is created by accumulating motion areas detected in an image sequence by the frame differential method, changing intensity of them for every frame. The proposed method is robust to overlap of vehicles because it estimates speed of vehicles without detecting and tracking vehicles individually. At first, Temporal Template is created from accumulating frame differential results. Next, median intensity is calculated on each horizontal scan line of Temporal Template. And then, the relation between the vertical axis of Temporal Template and median intensity is approximated as straight lines. Average speed of passing vehicles is obtained from the inclination of these straight lines. Temporal Template contains no motion area if there is no passing vehicle or all vehicles almost stops on the road. Therefore, if Temporal Template has no motion area and the background subtraction image has no foreground area, our method judges that no passing vehicle exists on the road. On the other hand, if Temporal Template has no motion area and the background subtraction image has some foreground areas, it judges that all vehicles almost stops on the road. Experimental results including the scene of heavy congestion show the effectiveness of the proposed method.