A 3-D detection method of moving traffic loads based on the stereo vision

Vehicle tracking based on computer vision is an important part of traffic load monitoring. The stable and reliable vehicle detection is the primary task of vehicle tracking. The widely used 2-D detection works accurately in simple situation but faces obstacles in dealing with multi-vehicle overlap. Therefore, this paper proposes a 3-D detection method of vehicles based on the stereo vision, which can position accurately the vehicles partly covered in the image to keep a continuous vehicle tracking. This method builds spatial relations between observed objects such as wheels and the feature points on the vehicle. With spatial relations determined, when the observed objects of the vehicle are partly covered by other objects, their positions can be inferred accurately from the visible feature points of the vehicle. Experiments were taken to verify the effectiveness of the proposed method.