Traffic Flow Statistics Based On Vision

With the improvement of living standards, the increasing number of vehicles caused traffic problems, for example traffic jams, traffic accidents and vehicle exhaust pollution to air and so on. The intelligent transportation systems is full use of existing transport infrastructure and combine different disciplines to manage traffic, effectively easing the traffic pressure. Vehicle flow statistics provide the basis for decision-making data for the intelligent transportation systems, contribute to traffic management department of transportation optimization scheduling, help the driver better alternative travel routes, city planners may be made widening planning according to whether a road traffic parameters, it has a very important theoretical and potential value to research traffic flow statistics. Build the observation matrix for H, S, V three components, established RPCA model,by low-rank components optimized by three components of low rank portion and sparse portions initially been moving vehicleThis article is studies traffic flow statistics based on vision for road test electronic monitoring system, the main contents include: moving object detection and vehicle statistics. A algorithms of moving object detection is proposed, which is based on RPCA in HSV color space, vehicle statistical methods based on dual virtual detection line is designed, improve moving object detection and vehicle counting accuracy and robustness, provides accurate and reliable traffic parameters for intelligent transportation system. The main innovations are as follows:Firstly, since it is difficult to identify the presence of shadows, and low detection accuracy when moving object detection based on gray-scale information has poor ability, there is a new algorithm of moving object detection based on low-rank matrix decomposition in the HSV color space is proposed, the approach build the observation matrix for H, S, V three components, establish RPCA model, obtain the low-rank and sparse of three components by low-rank matrix optimized, through which moving vehicles can be initially separated from the video; noise removal and cavity filling for the moving vehicles obtain accurate foreground moving vehicles. The experimental demonstrate that the proposed method is much better than others in increasing accuracy of moving objects detectionSecondly, Depth study of commonly vehicles statistical methods: target tracking method and virtual coil method, A novel method is proposed based on dual virtual detection line. Firstly, set up a dual virtual detection line to form a virtual detection region.Secondly, the vehicle information is extract from the virtual region, transform the vehicles information into a one-dimensional function, amend the vehicle information. Thirdly, design vehicle counting rules. Finally, vehicle counting if the vehicle disappear the virtual area.Different road traffic videos are used to verify the proposed methods. This experimental results show that the moving object detection algorithm can improve the accuracy and robustness, the result of the traffic statistics obtained by using the method has higher accuracy and feasibility.