Study On The Intelligent Recognition Of Video Surveillance In The City Tunnel

Intelligent Transportation System is the current subject of worldwide extensive research. In nearly 20 years, with the rapid development of Chinese road transportation and the increasing of Chinese long road tunnels and underground channels, the tunnel's own traffic monitoring and security management have become a new research topic. Application of ITS technology to monitor the tunnel is better than the traditional sense coil detection method. The technology of using video to control traffic has the advantages of easy installation and inexpensiveness.The common video vehicle detection technology is mostly used in open-air vehicle control test. While the road tunnel is a special section of the tunnel in which it has narrow space, poor air circulation, dim light, poor visibility and other special circumstances, which are the potential threat to traffic accidents. In different situations, the technology of video-based vehicle detection is different. On the basis of previous studies, this essay summarizes a proper tunnel monitoring, robust, real-time approach to deal with video images of traffic provided by a range of cameras which are fixed in the tunnel so as to detect vehicles, classify and track vehicles within the surveillance. The algorithm mainly includes three aspects:moving object extraction, vehicle identification classification and vehicle tracking.The algorithm of extracting the moving object, by comparing with several traditional algorithms, summarizes to use the background difference method to extract moving objects from the current monitoring region. Background subtraction consists of two steps of background modeling and threshold difference.Then we classify the moving targets according to the requirements of the tunnel for vehicles, using the BP Neural Network. After consideration of the various features of moving targets of the tunnel carefully, moving targets are proposed as input under the moment, speed and area features; classification categories as output. The neural network is trained and verified the feasibility.In real-time moving object tracking, the tracking algorithm based on KALMAN filtering is the most common and effective way. However, in the tunnel monitoring, it often appears many objects crossing or overlap situations, which easily lead to the phenomenon of the loss of target tracking or wrong target tracking. Therefore this essay uses a tracking algorithm based on a feature point matching with KALMAN filtering. Experiments show that the algorithm can effectively improve the recognition rate of multiple moving vehicles in order to achieve the right tracking.Through a large number of experiments on the video images in the tunnel, the results of the method in this essay show that:the method is real and valid.