Research On Traffic Flow Detection Algorithm Based On Video Image Processing

Intelligent video surveillance is an emerging and leading research subject in computer vision field, which can automatically analyze the input image sequences, and make reaction decisions according to the results of moving object detection, tracking and behavior understandings. Compared to the traditional traffic surveillance system, intelligent video surveillance system has saved a lot of manpower and material resources and been more accurate. It had been wildly used to monitor in military, traffic, bank and some other important places. Meanwhile using video technology to get traffic data is the main and key skill in intelligent transportation systems, which can cover large area and obtain rich traffic information.The thesis mainly focused on three parts:moving vehicle detection, vehicle's self moving shadow detection and removal, and finally traffic flow information acquisition. In the moving vehicle detection part, the author firstly made some research on several common used moving object detecting methods, and then proposed a new method combining image sequences histogram and single Gaussian to get the real time background. The proposed method is easy and can quickly get an accurate background, which is easier and faster than Mixture of Gaussians with similar background effects and can handle slowly moving objects. At last, the OTSU method is selected to get the binary images, which can get the best experimental results with most moving vehicle information. In the shadow detecting part, after analyzing advantages and disadvantages of color-based method and gradient-based method, the thesis put the two methods together to detect more accurate shadow area and maintain more complete vehicle information. In the last part, virtual loop method is introduced to detect traffic flow. In the traditional method, square coil area is the main reference of the detection, but here a more accurate reference is put forwards using the centre point of square coil, which can effectively reduce the effect of vehicles occlusion and running between lanes. Traffic flow detecting algorithm based on video image processing can well realize its expected functions with easy principle, fast computation and accurate acquired traffic flow data.