Traffic Detection Algorithm Research Based On The Video

With the vigorous development of the urban transportation in our country, vehicle ownership is greatly increased in the city. The development of the transportation system has become an important factor of urbanization. Traffic detection based on the research of computer vision has become a focus and hot topic at present in the intelligent transportation systems.The main content of thesis paper is to extract information of moving vehicle and remove its shadow in the video image to get whole object, obtain the real-time traffic parameters. It’s focused on three parts: moving target detection, shadow removal, and traffic-flow statistics.In the first part, an improved moving target detection algorithm is proposed based on the combination of three-frame difference and background difference according the problems that target-moving detection is affected by the background and inter-frame difference is easy to produce hollow. The initial background is established by edge detection and averaging method, and the video images are divided into multiple sub-block. Adaptive threshold is carried out in the sub-block of image based on the improved method, and foreground moving target is acquired with improved adaptively method to background image. This method is effective to extract the motion area and noise pollution is reduced, has a good robustness.In the second part, an algorithm is proposed for removal of moving shadow based on YCbCr color space and texture features with disadvantage of shadow removal methods utilizing YCbCr color space. The shadow area is determined by analyzing the color statistical properties of the difference between the foreground and background areas in YCbCr color space, then the improved LBP operator is used to segment the shadow area and remove the shadow precisely.In the last part, an algorithm of virtual-window traffic flow detection is introduced to determine vehicle-passing or not through changes of covered statistical window.The algorithm proposed is simple based on video traffic detection, and can be more accurate to extract the traffic parameters. Actual traffic video is tested to show the accurate number of cars.