Vision Based Moving Vehicle Detection And Tracking

With the development of city and economy, the burden of the road transportation system becomes higher and higher. Therefore, more attention has been paid to the vision-based intelligent transportation system. The key problem of this application is vehicle detection and tracking from the video sequence captured by static traffic surveillance camera. In this paper, we do some researches on the key technologies of vehicle detection and tracking, which play important roles in the vision-based intelligent transportation system. Our research works are summarized as follows.Firstly, we propose Gaussian motion model for moving vehicles segmentation in the dynamic scenes. By investigating the distinction between the motion vectors of the dynamic background and those of the moving vehicles, it is found that the motion vectors of the moving vehicles cluster in a small region while those of the dynamic background are dispersive. Consequently, Gaussian motion model is proposed to model the motion of the moving pixels in the scene, and Bayesian framework is employed to classify moving pixels into the moving vehicle or the dynamic background. At the same time, we make use of the lighting change of the scene and propose an adaptive background learning rate, by which the background can be updated adaptively according to the lighting conditions.Secondly, ratio edge is employed to detected moving cast shadows. Ratio edge is calculated as the ratio between the intensity of one pixel with that of its neighbouring pixels, and we can prove that ratio edge is illumination invariant. The background subtraction is then implemented in ratio edge domain. The distribution of the normalized background difference of ratio edge in shaded background area is analyzed and is approximated to be aχ2-distribution. A significance test is then used for automatic shadow detection. Intensity constraint and geometric heuristics are utilized to enhance the detection results, and an iteration strategy is implemented to estimate the shadow intensity ratio, which is defined as the ratio between the intensity of the shaded image and that of the background image. Experiments are conducted on various typical scenes, and the results exhibit that the proposed method can detect moving shadows robustly. Quantitative evaluation and comparison demonstrate that the proposed method significantly outperforms state-of-the-art methods.Thirdly, we present a multilevel framework to detect and suppress vehicle occlusion. The proposed framework consists of intra-frame level, inter-frame level, and tracking level. In intra-frame level, occlusion is detected by evaluating the ratio between the area of vehicle and the area in the vehicle's convex hull, and the detected occlusion is suppressed by removing a"cutting region"of the occluded vehicles. In inter-frame level, occlusion is detected by performing Lilliefors test on the motion vectors of the detected vehicles and the occluded vehicles are separated according to a binary classification of motion vectors. In the tracking level, an occlusion layer model is constructed and maintained adaptively, and the detected vehicles are tracked both in the captured image and in the occlusion layer images by performing a bidirectional occlusion reasoning algorithm. The proposed intra-frame level, inter-frame level and tracking level are implemented sequentially in our framework. Experiments were conducted on various typical scenes and the results show that the proposed framework can detect and suppress occlusions effectively.