Key Techniques Of Traffic Moving Objects Detection Based On Machine Vision

With the rapid development of Chinese economy, the scale of urban expands unceasingly. The rapid increase of urban population and motor vehicle ownership lead to the traffic problem increasingly prominent. The role of intelligent transportation system(ITS) in traffic management and traffic parameters acquisition is more and more important. Intelligent transportation system based on video has the advantage of convenient maintenance and installation, low cost and wide detection range. So it becomes one of the important ways to obtain the traffic information. How to detect and track the vehicles and pedestrians effectively in complex traffic scenes is one of the core functions of ITS based on video. So it is of great theoretical value and practical significance to take the moving vehicles in traffic scenes as the research object of target detection and tracking.Some key issues of the traffic objects detection and tracking based on video are studied deeply in this paper. First, this paper introduces the commonly used methods of image enhancement and morphological processing in traffic scenes. Then, this paper researches the background difference methods based on the Gaussian mixture model deeply and proposes an improved background modeling method for the traffic scenes with high occupancy of road. The improved method combines the sequence average algorithm and the Gaussian mixture model algorithm, updates the Gaussian mixture model selectively. The improved method can establish more accurate background model and reduce the amount of calculation. The advantages of the improved algorithm are particularly evident in the traffic scenes with high occupancy of road. For the problem of vehicle shadow elimination, this paper puts forward a method based on normalized RGB space. The experimental results show that the method has a good effect of shadow detection. In multi-target separation link, this paper researches two typical connected components labeling methods and puts forward a new connected components labeling algorithm which can effectively reduce the probability of labels conflicting, thereby reduce the algorithm’s time complexity. In the part of traffic object tracking, this paper researches the principle of Kalman filter and Mean Shift tracking algorithm and analyzes their characteristics by experiments. Then, this paper puts forward a multiple targets tracking method which combined with the characteristics of traffic scenes and made full use of the advantages of the two algorithms. The experimental results show that the improved algorithm can achieve a good tracking effect in complex traffic scenes and has high value in engineering.