Research And Application Of Vehicle Face Recognition Technology Based On Deep Learning

At present,there are some problems in the intelligent parking fee collection sy stem based on license plate recognition technology in China,it is unable t o identified unlicensed vehicle that could reduce the efficiency of intelligent parking system,and evading fees from fake plate vehicle that could infringe the rights of the legitimate owners.To sum up,It is necessary to use other vehicle features besides license plate to accurately identify vehicle identity information,vehicle face has rich vehicle feature information,therefore,we use deep learning technology,the vehicle identity info rmation is effectively identified base on vehicle face image features,and then the inte lligent parking charge system is designed and implemented,it solve the problem that the unlicensed vehicle cannot be identified and fake plate vehicle evades fees.Vehicle face recognition technology is a kind of technology to identify and a uthenticate the vehicle according to the front characteristic information of the vehicle.With the development of science and technology,vehicle face recognition technology has been widely used,such as vehicle face identification,vehicle face verification,etc.In the process of vehicle face recognition,it is often accompanied by vehicle face attribute analysis,the vehicle face attributes are obtained through algorithm analysis,such as vehicle color,vehicle model,etc.The vehicle face attributes are helpful for us to make further judgment and application in the actual scene.The main content of this thesis includes the following four aspects:(1)Research on the algorithm of vehicle face detection.The network structure of yolov3 is optimized to realize fast and accurate detection of vehicle face in v ideo.(2)Research on vehicle attribute analysis algorithm.VGG network structure is optimized to extract vehicle face image features,identify the vehicle color and model through softmax function.Compared with other typical vehicle color or model recognition algorithms,the algorithm has higher recognition accuracy.(3)Research on Algorithm of vehicle face recognition.The deep residual network structure is improved to extract the features of v ehicle face image,and the tripplet loss function is improved to train the network model.The experiment results show that the recognition accuracy of vehicle face recognition algorithm is 94.32%..(4)Research and development of intelligent parking charge system.We have developed a set of intelligent parking charging system through above research results.The charging system solves a lot of problems,such as the unlicensed vehicle can not been identified,the fake plate vehicle escape fees,etc.The charging system provides "vehicle face payment" function,it provides an example for the innovative application of parking fees.