A Study Of Vehicle Detection And Tracking Based On Vision

As a core part of the traffic monitoring system, vehicle detection system has a veryimportant research value. Vehicle detection system is divided into day and night detectionsystem. At present, the research on the day vehicle detection system has been studied verythoroughly and gradually improved. However, the day vehicle detection method can hardlyused at night because environment relatively dark, less objects to refer and performanceinstability. At present, under the scene of nighttime vehicle detection, we can’t handle thereflected light on the road, so the rate of the vehicle detection decreased. In order to solvethese problems, this paper focuses on the method of eliminating reflected light and researchthe vehicle detection and tracking technology. The main researches are described as follows:1) In the problem of too much interfere information at night, this paper proposed aclassification method which is based on Fisher discrimination method. This method is apretreatment method which can remove the reflected light. First, this article extract the regionof interest, this method can reduce the amount of computation and error. Then, this paperusing the method of histogram twin peaks to distinguish foreground and background lightseffectively, because there are obvious different between foreground and backgroundcharacteristics at night. Finally, this paper proposed a Fisher discrimination Analysis based onthe gradient of the region difference between vehicle lights and reflected light. Combinedwith atmospheric scattering and Laplacian of Gaussian, this paper used Fisher discriminationmethod to remove reflected light.2) For nighttime vehicle detection, this paper proposed a vehicle detection method whichis based on the headlights. First, extract the effective area of headlights through morphologyto eliminate some area out of the headlights. Then based on the prior knowledge that thevehicle is form by a pair of headlights and compared the distance between the two headlights,their symmetry and shape, then match the two headlights. If the match is successful, the twoheadlights as the same vehicle are detecting and tracking. Finally, the unusual circumstancesare discussed and treated, thereby improving the correct rate of nighttime vehicle detection.3) For the phenomenon that the videos are made up of multi-frame images in the sceneof road traffic, this paper proposed tracking method which is based on neighborhood. Thismethod is combined with the pre-track queue and stable tracking queue. The pre-trackingqueue is used to sift out the stable tracking target which can enhance the stability of tracking algorithm. The stable tracking queue is used to enhance the fault tolerances of losing targetand changes of the target features. This method can effectively reduce the losing target due tothe interference of background light, thereby greatly enhanced the anti-jamming and theaccuracy of tracking methodThis paper using Visual Studio2008programming method can detect and track thenighttime vehicle in the Windows system. The experimental results of urban road videos showthat the proposed algorithm has good stability, accuracy and real-time performance.