Research On Video Vehicle Detection Technology In Intelligent Transport System

Intelligent traffic video monitoring system(ITVMS),using image processingtechnology to analyse video images,including vehicle detection,vehicle tracking andlicense plate recognition three processing modules.Among them,vehicle detection is oneof the key technologies of ITVMS, whose main task is to segment vehicles from videoimages.As the premise of vehicle speed detection,vehicle behavior analysis,license platerecognition and other functions,its importance is obvious.Vehicle detection based on thebackground substraction has been widely used since its simple algorithms and it will notproduce hollow.Currently,most of algorithms research under the condition of video images during theday,while algorithms of vehicle detection are relatively less,this is because the nightenvironment is not only based on the complexity of daytime environment,but increasesmore complex factors such as street lamps and other lights’ illumination change, stronglamp light, pavement reflection and so on.Therefore,this paper aims to use differentvehicle detection algorithms for day and night environment,which can real-time andaccurately segment vehiclesunder circadian conditions.The paper introduces four steps ofthe vehicle detection process in turn: video images preprocessing,background modelingand updating,foreground detection and adhesion segmentation,this paper mainly asfollows:（1）For the daytime traffic video,firstly we take advantage of the mean dropsampling for pretreatment,then use the gaussian mixture model for background modelingand updating,we obtain foreground images through the integration of results of color spaceand grayscale gradient space background subtraction.（2）For the nighttime traffic video,we use mean drop sampling,gray leveltransformation and logarithm stretching for pretreatment,then set a certain size of slidingwindow In the time axis,and use median filtering method on gradient video images forbackground modeling and updating,finally obtain foreground images through thegrayscale gradient space background subtraction. （3）After morphological filtering and connected region analyzing,we judge whethervehicle areas are sticked or not through the Freeman chain code in foreground images,thenuse Graham scanning method to detect convex hulls, realize the segmentation of adhesionvehicle areas.（4）Completed the traffic light color recognition algorithm in video images,which isbased on the fast transformation formula of HSV color space.We finish a vehicle detection system based on the above work,through the testingsamples of day and night traffic videos,which can real-time and accurately segmentvehicle areas,and have traffic light color recognition, red light violation capture andvehicle flow statistics functions.