Research Of Pedestrian Detection And Tracking Algorithm In Intelligent Transportation System

With the rapid development of urbanization, the urban traffic pressure increasesrapidly and traffic accidents frequently happened. So the research and development ofintelligent transportation had received more and more attention. Intelligent transportation isa huge system, in which the pedestrian detection and tracking based on machine vision is avery important content. It can be applied in the intelligent Driver Assistance System toreduce traffic accidents and casualties. It also can be applied to intelligent transportationsystem for the statistics of pedestrian traffic and the analysis of abnormal behavior. So thepedestrian detection and tracking has great practical significance.This paper researches and compares the existing pedestrian detection and trackingalgorithm, analyzes the advantages and disadvantages of them. In order to describe thediversity of human appearance and contour, and solve the randomness of the humanmovement, this paper combined the histograms of oriented gradients (HOG) feature withsupport vector machine (SVM) to complete the pedestrian detection. Experiments showthat this algorithm can accurately district non occluded branch and other target orbackground. Due to the computation of HOG is time-consuming, this paper uses themethod of integral figure calculate the HOG to improve pedestrian detection speed. Inorder to describe the pedestrian more accurately, this paper uses the combination the localbinary pattern (LBP) with HOG to detect pedestrian, this method significantly improvesthe pedestrian detection rate. Because the vehicles and buildings often block pedestrian,this paper uses the pedestrian body classifier and bust classifier, which trained with theINRIA database and own pedestrian pictures. Experiments show this method can improvethe detection effect when the target is blocked.In terms of pedestrian tracking, this paper completed the pedestrian tracking usingparticle filter method with the color feature and the HOG features. When the newpedestrian appeared in the video, this pedestrian will be joined the detected ranks by detecting the images periodically; therefore enhance the robustness of tracking withconstantly updating the pedestrian. The experimental results show that this method canovercome the defects by using single color features to track, reduce lost phenomenon whenthe background color close to the pedestrian or color change of pedestrian, improved therobustness and tracking accuracy.