Preceding Vehicle Detection In Bus Lane Based On Bus-video

The rapid development of road transport provides a great convenience forpeople's living, but with the growing number of all kinds of cars, some illegal actssuch as other cars occupying bus lanes and bus stations become more and more,whichseriouslyimpacttheefficiencyofpublictransportaswellastheentiretransportsystem's smooth. Fixed electronic police is difficult to monitor these illegal actsabove because of its inherent defects; intelligent monitoring system mounted on busprovides bettersolutionstotheproblem.Forthebus lanemonitoringsystem mountedonthe bus, howtoeffectivelydetectthefrontbus laneandaccuratelydetect thefront cars inthelane is veryimportant ,asonly detected the vehicles, can the vehicles whether illegally occupy the bus lane bedetermined. As the front vehicle in the video is very near from the bus, the red taillights of the vehicleare very obvious,this paper just uses the red characteristic ofthevehicle's tail lights to detect the vehicle. The First step is to extract vehicle candidateregion in the HSV color space,then verifythe candidate region to confirm whether itisthe realvehicle target. The main content and innovation of this paper are providedasfollows:(1) The lane detection. The paper first extracts the straight lines from the videoimage based onHoughtransform,then based ontheestablishedroad model,proposeslane extraction algorithm, the algorithm extracts the driveway from the chaotic linesgot by Hough transform. And based on the extracted driveways, the road region candedeterminedwhichrestricts thedetection regionforthevehicledetection.(2) Vehicle candidateregion detection.This paper proposes the vehicle detectionmethod in HSV color space based on the monitor system's requirements for thedetection. First , convert the video image to HSV color space from the RGB colorspace,and extract the Huecomponent inHSV colorspace,thendetect thetaillights from the Hue component image, by which the vehicle candidate region can belocated.(3) The verification of candidate vehicle location. There often are some faketargets in the candidate vehicle location, so it is necessary to identify real vehiclesfrom those hypotheses. Verifying a hypothesis is essentially a classification problemof two patterns, i.e. vehicle and non-vehicle. By applying machine learning, certainamount of vehicles and background samples are extracted. Then this paper uses GaborFilters feature extraction method and applies Support Vector Machine (SVM) to trainthe classifier. Firstly, extract features from the samples and put the feature vectorsinto SVM to learn the decision boundary. During the verification of candidate vehiclelocation, feature vectors are extracted from those regions and put into the classifierwhich has been trained to get the classification result. And finally false targets will beremoved and target vehicles are got.Experiment results prove the practicability and validity of the method for the lanedetection and the method for vehicle detection based on HSV color space. And theverification experiment based on Gabor filters and SVM also proves that theclassification method can effectively improve the accuracy of vehicle detection.