Research Of Video Traffic Flow Detection System

Video traffic detection systems, which belong to the intelligent video analysis application areas, are the indispensable components of intelligent transportation systems. By installing cameras on the road to get traffic video source, the traffic flow detection system detects the number of vehicles in real-time statistics, and outputs the result to the terminal management, which provides an important basis for traffic management and decision-making, plays an important role in urban traffic control. Because of traffic detection technology is not mature, especially serious error detection, missed detection, and poor operational efficiency can be emerged in complex background. Guided by the existing research results, the further research and analysis were discussed in this paper. Focusing on the method of vehicle detection, the comparison and testing indicate:Based on Haar features and the Adaboost cascade classifier vehicle detection algorithm can achieve good results applying to traffic video detection system. The algorithm can greatly conventional vehicle detection methods to improve the inevitable condition like error detection and missed detection, especially in complex background. The main work is as follows:(1) Video capture and video preprocessing. The installation of the camera and the traffic video capture were simply described, and the image preprocessing methods were studied according to the actual application. The results show that:median filter can restrain the salt and pepper noise, and histogram equalization is the most widely applicability image enhancement algorithm.(2) Vehicle detection. Several common vehicle detection algorithms based on image processing were analyzed in general, and the advantages and disadvantages of each method were summarized, where the background subtraction method was mainly analyzed. The vehicle detection algorithm based on haar feature and Aaboost cascade classifier, which are from Viola and Jones face detection thoughts and were introduced to vehicle detection by researchers a few years ago, is mainly studied. Finally, the overall effect comparison between this detection method and background subtraction, concludes:This algorithm is simple, robust, has good detection results and fast test speed, especially in a complex background.(3) Track count. Several common tracking algorithms are analyzed, and combining with specific practical application, choose the tracking algorithm based on region for the detected vehicle, and then complete the vehicle counting. The results show that:this method can achieve very good vehicle tracking, and help to further complete the vehicle count.(4) System implementation. Haar features combined with Adaboost cascade classifier vehicle detection algorithm were applied to the construction of traffic detection system, and use the area track algorithm on the detected vehicle to complete the traffic count. Use Microsoft Visual C++2010programming environment under Windows XP system, and then achieve the function by calling the Opencv library. After testing and validation, the results show that:the performance of traffic detection system is good, whose accuracy rate is about90%, and can meet the real-time requirements.