Recently, since the traffic is increasing every year and the problems of traffic jam and accidents, and so on, are increasing, it is required that the Intelligent Transport Systems (ITS) provide the drivers with the information for smooth and safe driving. As one of ITS techniques, the studies on the traffic flow measurement by image processing have been proposed widely. In the conventional methods, since the traffic flow is almost always observed from one viewpoint, thus one vehicle is occluded by another one. It is difficult to extract an occluded vehicle by the image processing. In the authors' proposed method, multiple views are used in order to solve the occlusion in the image taken from one viewpoint. On the other hand, the location of cameras are not always placed at the desirable position for the traffic flow observation system. The authors propose a method to extract the movement vehicles from the multiple viewpoints which are placed as freely as possible. Even if the occluded vehicle exists in the image taken from one viewpoint, the authors' method solves the occlusion by using the image from another viewpoint.
Abstract