Vehicle Detection For Anti-Occlusion Under Complicated Background

With the rapid increase in the number of vehicles, How to manage vehicles effectively to reduce the rate of accidents has increasingly become the major problem many countries pay attention to. In view of this situation, all kinds of intelligent transportation systems(ITS) are proposed at the historic moment, and vehicle detection is the foundation of the system.At present, optical flow method, the frame-difference method and background subtraction method are commonly used in vehicle detection technology. Among them, background subtraction method has better performance in real-time and accuracy if a good background model is established.In this paper, we use modified mean-value filtering method to establish background model. The modified method gets higher accuracy than the traditional method in the complex background by bringing in the histogram threshold value. At the same time, because of the influence of illumination, the shadow is detected together with the foreground vehicle, which has a serious impact on subsequent tracking and analysis. So we need to remove the shadow. In this paper, we use modified shadow detection algorithm base on HSV model, and this algorithm enhances noise immunity of single pixel point by bringing in eight neighborhood mean-value. Moreover, because of the serious vehicle congestion situation, the mutual occlusion among vehicles is inevitable, how to separate the vehicle occlusion is the necessary step after vehicle detection. In this paper, we put forward anti-occlusion algorithm based on angular point. Firstly, we pick up angular points through discoid moving algorithm, secondly, we store the angular points in a circular linked list, thirdly, we compare the data in circular linked list with the template library and find out the matching template, at last, we conduct the occlusion separation of the vehicle according to the known matching template. This anti-occlusion algorithm based on angular point can be applied to all kinds of common occlusion, and the algorithm has good feasibility.