Research On Vehicle Tracking And State Identification Method For Parking Detection In Highway Tunnel

Due to the sealing of the scene and the high driving speed, if parking accidentscannot be found and treated timely, will easy to cause secondary accidents, even leads tomajor traffic accidents, impacts on the operation of the highway seriously. Video-basedmethod is the major way of parking detection, but the main study is based on simplescene. While the actually highway tunnel environment is more complex, the existingparking detection methods cannot satisfy the actually needs. Therefore, according to thepresence of light changes, vehicle light interference and video image blur in highwaytunnel, giving full play to studying the technology of parking detection, to realizing theparking detection quickly and accurately has important practical significance.The presence of light changes, vehicle light interference and video image blur inhighway tunnel, bringing great difficulties to vehicle detection, vehicle tracking andparking identification.For these complex cases, this paper studies the modeling andupdating of background, selection and matching of features, the selection of the initialtracking target and the identification of parking target and proposes the treatmentmethods.In the aspect of vehicle detection, selects the background modeling method basedon nonparametric kernel density.Considering background overall mutations, backgroundchanges partly and parking these situations, this paper presents a multimodalbackground updating method. The experiment results show that the backgroundmodeling and the background updating method can effectively modeling thebackground.In the aspect of vehicle tracking, choices and extracts the characteristics of color,shape and location, and does multiple features fusion by matching method.In addition,according to problem which the change of tunnel environment will changes the featuresvalues of comprehensive characteristics, proposing a kind of method of self-tuning offeature’s values. For highway tunnel is a complex environment which exists target lostand disturbance, proposing the mothod of choice of initial tracking target based onvirtual detection line. In order to excluding the pseudo prospects caused by partialillumination changes and the vehicles which unlikely happening parking event from thetracking sequences, proposing the mothod of treatment of the tracking process. Theexperimental results show that the proposed method of self-tuning of feature’s values can update the feature weights automatically and match the actual environment moreaccurately. The proposed method of the choice of initial tracking target and thetreatment of the tracking process can ensure the accuracy and reliability of vehicletracking process.In the aspect of parking identification, on the basis of vehicle tracking, this paperproposed a classification of the parking target recognition method which doing thepreliminary parking identification based on the number of vehicles tracking successfullyand the doing the final parking identification by the change of the coordinates of thecenter of vehicle mass. Experimental results show that the proposed method can quicklyand accurately identify the parking target.Finally, combining the above methods of each part, using the surveillance videodata from the Chongqing highway tunnel, this paper did experiments in the VCenvironment. The results show that the proposed method can extract and track thevehicle, the parking identification method can quickly and accurately identify theparking target.