Research Of Intelligent Algorithm In Highway Tunnel Video Surveillance

In the past few years, the intelligent transport system has been developing byleaps and bounds. By virtue of the realization of intelligent management of the trafficsurveillance process, not only the ITS saves the manpower and material resourceseffectively, but also improves the efficiency and security of transportation greatly, andconsequently brings about a noticeable effect on the growth of national economy. Theresearch on the traffic intelligent surveillance and management that is based on thecomputer visual technology is a new hot spot in ITS. This method realizes vehicledetection and tracking processes quickly and effectively by means of imageprocessing technology, and has an advantage of low cost and convenient maintenance.Its practical and economical feature has received a wide admission in this area.Tunnel scenes commonly exist in the highway and urban roads. As the bottlenecksection and the accident black spots of transportation, the highway tunnel scenesconsists of almost all kinds of complex conditions and has a great value of researching,which makes a great challenge to the implement of traffic surveillance policies. Thispaper is dedicated to analyzing and studying the video surveillance algorithms in thehighway tunnel scenes, specially the vehicle detection, location and trackingalgorithms.Firstly, this paper clarifies the subject background and researching significance,gives the definition of ITS and review the historical development, and analysis thecharacteristic of highway tunnel scenes and complications of the design ofsurveillance algorithms.Secondly, this paper elaborates moving object detection methods, object trackingalgorithms and their adaptability, and introduces the feature subtraction methods ofmoving vehicles. Of this total, we mainly analyze the background modeling problemsin moving object detection, and propose a delay-time background updating modelbased on single Gaussian distribution which is adaptive to the suddenly-changedbackground and gradually-changed background.Thirdly, this paper studies the vehicle detection in highway tunnel scenes with streetlights. Especially we clarify and analyze the methods of vehicle subtractionbased on texture feature segmentation methods. As to the low adaptability of normalmethods to illumination effects, we propose a moving vehicle subtraction methodbased on local normalization, which has a good effect on moving vehicle subtractionin tunnel scenes with streetlights.Lastly, this paper studies the vehicle location and tracking methods in highwaytunnel scenes without streetlights. We propose a highly-robust vehicle detectionmethod from the headlight angle which obeys the idea of “Tracking firstly, clusteringsecondly”. This method makes the most of temporal and spatial information ofmoving objects and realizes exact vehicle location and tracking process. The goodveracity of this method in tunnel scenes without streetlights is proved in thesimulation experiments.