Research On Moving Vehicle Detection And Vehicle Flow Detection Method In Video

In intelligent traffic system (ITS), moving vehicle detection based on video played a very important part. However many factors, such as camera dithering, different light shades, complexity of traffic scene and the diversity of vehicle types, bringed a big challenge to the real-time property and accuracy of the detection. On the other hand, the collection some basic parameters of road traffic information, such as vehicle flow, type and speed was the premise and basis for the effectiveness of ITS. This paper presented a thorough study of electronic image stabilization algorithm, codebook algorithm and traffic flow detection method, which were the key points to solve these problems.Image processing methods in traffic video were also discussed. And preprocessing and morphology were performed on captured traffic images. Through experiments, the quality of commonly used algorithms was compared and analyzed. Given the video camera dithering problem caused by wind or truck passing by, the algorithm of electric digital image stabilization was studied, and grey projection algorithm was realized through program. Shaky removal experiment which is performed with test videos indicates the algorithm solves the problem about jiggle to some extent, achieving the goal of image stabilization.It was found that background modeling was a difficult point in vehicle detection. To solve this problem, codebook algorithm based on background difference technology was studied particularly. First, the algorithm established a codebook model based on YUV color space. Then a local updated layered codebook combining with frame difference was proposed, which reduced the detection time and adapted to the background changing. Besides, this paper researched the object detection method of local binary pattern (LBP), and analyzed LBP operator, then presented a uniform LBP algorithm based on picture block. Also the paper synthesized the properties of the two algorithms——low false positive rate of codebook algorithm and computational simplicity of LBP, first used LBP algorithm based on picture block to distinguish background blocks and foreground blocks, second used codebook algorithm based on pixel to detect pixels of foreground blocks. Experiments show that the new algorithm can guarantee the detection precision and improve the processing speed.In this paper, we designed and realized a traffic flow data collection system, which utilized virtual loop to calculate traffic flow and judge vehicle-type. The change of the number of pixels in virtual loop was used to detect a vehicle passing by or not. Vehicle type was judged by the passing time. The next step is to take some further research on traffic flow detection on the condition of trans-line vehicles.