The Research On Vehicle Reorganization In Complex Background

The thesis focuses on the vehicle recognition based on the traffic video. Vehiclerecognition is an important part of the intelligent transportation system. The intelligenttransportation system used to replace part of the people’s work or provide the information fordecision-making system. Vehicle recognition is a prerequisite of many subsequent processing.Compared with single traffic flow information, it has a greater reference value; it is thesupport technology for the electronic toll collection system.A new method of vehicle recognition is proposed, which depends on the apparentcharacter of the vehicle ceiling contour, and a contour-repair method is proposed to get aexact contour which is like the real contour, in the end. The type of the vehicle is recognizedby comparing the contour got in the frame with the models of different type of vehicles.First, if it is assumed that each pix of the picture is independent, and it meets theGaussian mixture distribution. A background model of the video can be established with theGaussian mixture model. Then the moving target mask could be got. If the pixel meets thebackground distribution, it will be classified as the background. The obtained binary image isconsidered as the foreground mask. Then connected region analysis can be used to get theposition and the bounding rectangle of the movement of vehicles in the video.Second, calculate the Haar features of the sample images, and a wheel classifier istrained with the Adaboost algorithm. The most significant character for vehicle is the wheelsof the car. Wheel detection helps the vehicle contour extraction. Because the traffic video isfilmed on the highway, and taken on the side of the road, so the wheel is the special structureof the vehicle. And the wheel in the view can be attributable to a particular car. At the sametime, the location and size of the wheels are good for the contour extraction than the vehicleimages alone. The classifier can detect the vehicle wheel in the original image and it obtainsthe position and size information of the wheel. Haar feature is very suitable for such "block"feature, and it works well.Third, extract the vehicle contour of the binary image. For vehicle recognition system,the contour information of the vehicle, particularly the upper contour of the vehicle issufficient to vehicle type classification, the premise is to obtain a complete and accurateprofile. Extract the contours of the vehicle the mask images to combined with thecharacteristics of the vehicle to make it more realistic for vehicle contour.Cut the contours ofthe vehicle based on the wheel position information to remove interference of the shadow of the vehicle. Then a method to repair the contour is proposed to make the obtained contourexactly like the real contour of a vehicle. Contour is the distinct character of different type ofvehicles.Finally, calculate the Hu invariant moments of the vehicle contour and match it withtypical vehicle contour models. The contour of the vehicle is one of the most significantcharacteristics of the vehicle characters. The difference between different vehicles of differentfunction is more obvious. Therefore, the contour of the vehicle is a good feature to distinguishbetween the different models. As the side vehicle features are good for vehicle recognition,which is used in this paper.The most closely matched model will be considered as the type ofthe vehicle. The match strategy has the advantage of accurate identification, expansion ofconvenience and flexibility.