Research And Implementation Of Appearance-Feature-Based Vehicle Recognition Algorithm

With the development of the economy and the society,smart city and intelligent transportation are necessary.As a result,the Intelligent Transportation System is a developing tendency of Transportation Management in the 21st century.In an age of"Big Data",video data,as a kind of untrusted-data,should be analysis and used well.Indeed,processing the video data from traffic surveillance cameras is an important part of the Intelligent Transportation System.Nowadays,the video of traffic surveillance can just be used as the evidence of vehicle tracking system;because video cannot be recognized automatic but need to be manual recognition.On the other hand,the vehicle tracking and management are basing the vehicle license plate recognition.Vehicle license plate recognition bases the images of traffic surveillance camera.As a result,in order to take a photo,the vehicles need to stop and wait.And sometimes the license plate cannot be recognized successfully,for the license plate is fake or be sheltered.Therefore,this method is inconvenient.In this setting,a new convenient and intelligence method which is used to recognize the vehicle in the traffic surveillance video is necessary.The color of a vehicle,the appearance of a vehicle and the type of vehicle can be used to recognize a vehicle.Basing these features of vehicle,a car can be recognized from the video data.This paper researches the features of vehicle,and come up with an algorithm to recognize vehicles.In this paper,in order to realize the vehicle recognition in video,vehicle appearance feature is summarized,which is the theoretical basis of the vehicle recognition;the commonly foundation algorithms in video processing algorithms is analyzed;the method of vehicle feature extraction and fitting is realized;and the vehicle identification algorithm based on vehicles' features is presented.This algorithm can be divided into three parts:video prospect target extraction,physical feature extraction of the vehicle,vehicle appearance feature extraction and matching.The part of video prospect target extraction can be divided into three steps:background extraction,background subtaction,image filtering,shadows subtraction.In the step of background extraction,the mean background modeling,the single Gaussian background modeling,the Gaussian mixture background modeling and the mid-value background modeling are compared.After background subtraction,use the difference algorithm extract the area of prospect target.Then,filter the images of target area.Finally,subtract the shadow.The part of physical feature,the method of the color identification of based on the Bayes classification are come up with.In the part of appearance feature,some kinds of edge detection algorithms are compared,the method of appearance feature extraction and matching are come up with.In this paper,the experiments are developed in Windows system,and used OpenCV and C++ in addition,which show than the algorithms can detect the different vehicle in the video and recognize them.Therefore,the algorithms can be used to help the management of traffic system.