The Design And Implementation Of Pedestrians Recognition And Tracking System Based On Binocular Vision

As urbanization increasing, people’s social activities are becoming more and morefrequent. Therefore the research of passenger flow estimation has the important theorysignificance and the practical value. In the light of the application on intelligenttransportation and video monitoring, this thesis is largely concerned with recognitionand tracking of pedestrians based on Binocular Vision. The vertical view images can becaptured because of the restricted position of Binocular camera. In this kind of image,only the top of head in particular will be shown completely among all of the humanbody parts which can be difference from other objects. The information of heads invertical view image, the contour of top of head and the color features and texture of thehead region is from the vertical view image. Therefore, how to extract effective featurepedestrian’s head which is to distinguish from the other targets is the most importantpart of this subject.In this thesis, the main task of this thesis is summarized as follows, detection of tophead, stereo matching, target tracking. Firstly, the vertical view images is captured onthe restricted position of Binocular Vision device, the contour profile of top of head isquasi-circular. On the basis of analyzing two kinds of Hough transform circle detectionalgorithm principle and simulation experiments,21Hough that is gradient houghtransform better suits on the vertical view images as the main technique to extract thehead’s quasi-circle contour. At the same time, to further lower the time consume and toextract accuracy the head contour, near-extreme error is introduced in the accumulationprocess of applying the gradient operator to pick up the contour space, the factexperimental results show the proposed algorithm of21HT has high accuracy and goodreal-time performance. Following research of feature extraction, this thesis focused onstereo matching algorithm which is the key issue in stereo vision. Stereo matchingalgorithm which is four key components is the emphasis. On analysis of binocularcamera, this thesis improved the area matching algorithm which using fixed windowsmatching and new area matching algorithm based on disparity gradient, the stereo disparity of candidate region can be got relationship between head’s depth informationand geometrical characteristic is taken as identify targets by further. After headrecognition, tracking of human body on Binocular Vision uses extended kalman filterprediction on tracking combine with the radius of top head and the position of the centerbetween adjacent flames.At the end of this dissertation, recognition and tracking of pedestrians is designed.The experiment accuracy rate up to95%which be performed in the lab verify that thealgorithms for head of moving human body are useful and accurate.