Research On Pedestrian Detection Technology In Intelligent Transportation System

With the continuous development of science and technology level,technology in intelligent transportation system(ITS)more perfect,its application scenarios and demand is growing.Compared to the traditional intelligent transportation system,technology,a new generation of intelligent transportation system based on hardware or the software algorithm is increasingly perfect,and already have real-time timely high-quality performance such as high accuracy and low rate of false positives.A new generation of intelligent transportation system through the study of the all-weather,continuous monitoring of road conditions,as well as the real-time monitoring data processing by computer technology to high efficient calculation,as a basis for realizing intelligent road traffic.The pedestrian detection module as an important part of ITS system has become the core of the domestic and foreign scholars research subject,it have broad prospects for development in the field of scientific research and engineering applications.Pedestrian detection part of intelligent transportation system,this paper in-depth study and analysis,through the experimental analysis for high-definition cameras pedestrian images.Compared with the current popular pedestrian detection algorithm for experiments and in-depth analysis of improved algorithm is proposed in this paper.detection algorithm of job content are mainly concentrated in the following three aspects: first,Through image filtering,morphology processing and other related image processing technology,optimize the image sequence processing,retain the image on the basis of effective information less as far as possible the interference of redundant information;second,Selection of feature set,this paper selected different from the traditional feature set,single histogram of oriented gradient(HOG)feature and local binary patterns(LBP)features using statistical histogram cascade fusion,in order to get more deep more comprehensive image information;third,In part of the trained classifier to change the traditional way of training,it is using linear supportvector machine(SVM)classifier,adopting intercepting convolution neural network as classifier,the use of three full connection layers for training,finally get good classifier performance.At this point,it optimizes the real-time performance and robustness of the algorithm through the research and improvement of the above three aspects.It also reduces the pedestrian detection in the complex background of failure probability and improves the effectiveness of the pedestrian detection and reliability at the same time.