Research On Traffic Flow Detection System Based On Mean Shift Tracking Algorithm

With the rapid economic growth and the greatly improved standard of living, serious traffic problems occurred. Whether inland or abroad, traffic congestion has become a key affection factor contributing to the normal life of city residents. In order to improve this situation, national government has invested heavily in building roads and traffic infrastructure. But it is little effect. In recent years, as the development of computer technology, the intelligent transportation technology which can improve the poor traffic environment has gradually emerged. This technology has been widely applied in several foreign countries and the result is good. Currently, the intelligent transportation technology has been supposed as the best solution to the traffic issues.The intelligent transportation system makes the reasonable scheduling strategy to improve traffic environment, mainly through analysis real-time traffic information which fed back by the traffic flow detection system. The traffic flow detection system is the most critical link in the intelligent transportation system, which directly determines the final scheduling strategy of the intelligent transport systems. Therefore, this article is dedicated to the research of the traffic flow detection system. We first investigate and analysis the key technologies that will be used in the traffic flow detection system, then effort to improve its efficiency and finally accumulate experience.The traffic flow detection system consists of five parts: video capture, background extraction, vehicle detection, vehicle tracking and traffic flow parameters detection. This article not only gives a detailed description for every part, but also analyses and comprises the usual technical methods in each part. Chapter2 introduces several key technologies in the traffic flow detection system, which are background extraction technology, vehicle detection technology, vehicle tracking technology, etc. And the implementation processes and the practical effects of the technologies are presented.In the Chapter3, we first simply describe the algorithm processes of the traditional mean filtering algorithm for background extraction, and analyze its advantages and weaknesses by experiment. Although, the traditional mean filtering background extraction algorithm is simple to use, its effect is poor when dealing with the large traffic video stream images. In view of this, we propose a modified mean filter background extraction algorithm that can judge the video streaming images according to certain criteria, and then only count those who are background images, finally average they get the background image. In tests, it can be seen that the improved algorithm has overcome the shortcomings of traditional methods and adapts better when traffic is heavy.Chapter4 has studied the Mean Shift tracking algorithm in detail, and analyzes the advantages and disadvantages of the algorithm, also study the possibility to use the algorithm in traffic flow detection system. Because of its non-participation and robustness, the Mean Shift racking tracking algorithm has got more and more attention of staff. But the traditional Mean Shift tracking algorithm has high computational complexity, which is hardly to achieve the real-time tracking multiple targets need of the traffic flow detection system. So, this article give us the improved Mean Shift tracking algorithm based on linear prediction, which can improve the operating efficiency twice better than the original algorithm. Consider the traditional Mean Shift tracking algorithm can't change the nuclear bandwidth, the paper also proposed a nuclear-bandwidth adaptive update mechanism, which uses color information to determine the vehicles' current running state. The experiment result is very good.In the last part of this article, a software design of the traffic flow detection system is given. In the software design, we have integrated used the current mainstream technologies and the new methods proposed in the article. Finally, the availability and scalability analysis has been made for the system.