Research On Target Detection And Real-time Tracking Method Of Moving Vehicles

Moving vehicles are always the objects of study in Intelligent Transportation System at present.This paper proposes a method for detecting,identifying and tracking moving vehicles.In order to be able to identify different types of moving vehicles,such as large vehicle and small vehicle more accurately,a self-calibration method based on sample vehicle tracking frame and pixel ordinate of centroid of sample vehicle is proposed.Then,a definition of large and small vehicles is also given at same time.The method is used to prepare the standard for identifying moving vehicles and judging large vehicle and small vehicle.The specific steps are that the ratio between width and height of the tracking frame when small sample vehicle moves in a certain range will be obtained in advance as the basis for identifying the moving vehicle.Then,the width value of the adaptive tracking frame of the sample vehicle and the pixel ordinate value of the vehicle centroid will be obtained as reference for judging large and small vehicles.Subsequently,select a large sample vehicle,verify the identification and judgement to the large vehicle from the small sample vehicle by the relevant data of the large sample vehicle.After comparing the results of binarizing images from global adaptive threshold Otsu algorithm and the local adaptive threshold Niblack algorithm and Sauvola algorithm,the averaged filtering,median filtering and Gaussian filtering are used to denoise the generated binary image respectively.Then,the conclusion that the binarization effect based on the global Otsu algorithm is better than that based on the local threshold binarization method is obtained.Subsequently,aiming at the defects of Otsu algorithm that some vehicles' outlines are intermittent and unrecognizable after vehicle detection,give an improvement to it.After the superiority of improved algorithm is tested,the improved algorithm is selected as the adaptive binarization threshold method in this paper.In addition,based on the advantages of mean background method modelling fast and efficiently and the better quality of target foreground map obtained,mean background method is chosen for modeling the background in the paper.In the process of tracking a moving vehicle based on a connected area,a method of identifying the number of moving vehicles in a video image is also proposed.Then,Harris corner detection algorithm is chosen as the core of tracking vehicle method in the paper according to its stability and accuracy.The tracking method uses the center point of the feature corners distributing area as the basis for tracking moving vehicle and locates thevehicle body with an adaptive vehicle-sized tracking frame.According to the self-calibration method based on the sample vehicle tracking frame and the pixel ordinate of centroid,together with the definition of large and small vehicles,the target object is determined as a moving vehicle by the experiment accurately.Then,different types of moving vehicles are identified and verify the feasibility and accuracy of the method,proving that the related method in the paper is able to detect,identify and track the moving vehicles fast and accurately.