Algorithm Research Of Moving Vehicles’ Detection And Tracking Based On Traffic Video

With the development of today’s economy, the transportation industry develops rapidlyand the number of motor vehicle has increased rapidly. It is difficult to solve the trafficproblems (frequent traffic accidents, serious road press, etc) just relying on the road facilitiesand artificial management. Thanks to the traffic video, detection and tracking researches ofmoving vehicles’ based on these images can realize the automatic and intelligent managementin the traffic. These researches can make full use of the existing transportation facilities toensure greater transport capacity and improve the security of the transportation. Meanwhile, itcan achieve the intensive development of transportation industry.In this paper, detection algorithm of the moving vehicle is discussed based on the Gaussmodel, and tracking algorithm of the moving vehicle is discussed based on the Kalman filtertheory and Mean Shift iteration theory. The research results obtained are as following:Based on comparisons of the advantages and disadvantages of Gaussian single modeland Gaussian mixture mode, combined Gaussian mixture mode is built through edge pixel ofvideo image frame. Firstly, obtain a Gaussian distribution which can describe the model ofbackground pixels. Then, according to the matching relation to update the parameters of theGauss distribution. Finally, according to the matching relation of Gauss distribution and thepixels’ value to achieve the detection of moving vehicles.In Mean Shift tracking algorithm, since the kernel-band width is fixed, the changes ofvehicle within the kernel will not influence the localization accuracy of Mean-Shift trackingalgorithm, but it will cause enormous scale positioning error. So using the method of plus orminus ten percent bandwidth to updated kernel-band width updated automatically. Whenvehicle scale becomes larger, classic Mean-Shift tracking algorithm can cause scalepositioning deviation and spatial positioning deviation in the process of tracking. When thevehicles make zoom or translational motion within its window wide range, the spatialorientation of Mean Shift tracking algorithm is accurate. Firstly backward tracking method isused to make the centroid registration. Then, according to the matching corner of rectangular tracking box after centroid registration to establish affine model. According to affine model toget the stretching amplitude and updated the window width of nuclear.Since the Mean-Shift tracking algorithm can not effectively track the vehicles whichmove too fast, as a result tracking algorithm which combines the Mean-Shift algorithm andKalman filter is proposed. Kalman filter is used to predict the position of target in the currentframe, and Mean-Shift algorithm is used to get position of the vehicle.However when the Mean-Shift tracking algorithm can not effectively track the vehicleswhich are blocked heavily, Kalman filter model is closed firstly, then based on the position ofMean Shift point of convergence in previous frames, linear prediction is made in order to getthe possible starting position of the current frame, which means that the linear prediction ofcars’ location replaces the Kalman filter. Finally, using the Mean-Shift tracking algorithm toget the tracking position.Experiments show that a new moving vehicle detection algorithm which combinesGaussian mixture model and image edge information can detect moving vehicles in the fronteffectively. It can obtain the background in the case of light mutations by real-timebackground updating technique. Simulation experiment showed this algorithm could be veryaccurate and robustness when it is applied in complex environment outside. The method ishigh accuracy, robustness and has a wide range of practical. When the vehicle’s scale ischanged, an automatic bandwidth algorithm can not only effectively track the moving vehicle,but also the tracking window can change as the vehicle changes, which ensures the accuracyof Spatial positioning and scale positioning. When vehicle moves too fast or occluded heavily,the tracking algorithm which combines the Mean-Shift algorithm and Kalman filter can notonly track the moving vehicle effectively, but also can achieve the real-time detection.