Design And Implementation Of Vehicle Number Recognition System Based On Convolution Neural Network

Intelligent traffic system(ITS)is an effective solution to the current traffic management.As an important technical point of ITS,vehicle license plate recognition technology(LPR)has important practical significance for road traffic management and control.Researchers at home and abroad have been carrying out in-depth research on license plate recognition technology,mainly through the method based on image processing and template matching.Under the scene with clear image structure level,they have good adaptability,but largely rely on prior knowledge of image structure.This thesis analyzes and studies the four parts of the license plate recognition system,such as vehicle location,license plate location,license plate character segmentation and license plate character recognition.Based on the convolutional neural network technology,a complete vehicle identification system is designed and implemented,and the application service interface module of the system provides the vehicle identification function for third-party applications.The main work of this thesis includes:(1)After studying the methods of image target localization in recent years,it is found that the convolution neural network can be applied to image target detection based on R-CNN(Region-based Convolutional Neural Networks)target detection.To this end,this thesis proposes the r-cnn vehicle location and license plate location method,finds the license plate position through the target detection algorithm,and converts the target detection problem into the classification problem,so as to improve the accuracy of the previous license plate location algorithm based on image processing.(2)In this thesis,by analyzing and comparing the current mainstream methods of license plate character recognition,such as template matching,statistical classifier,and artificial neural network,it is found that image classifier based on convolutional neural network has deformation invariability to the target image.The space relation of convolutional network can be used to reduce the number of training parameters of classifier so as to improve the training ability.The license plate character recognition is realized based on the convolutional neural network AlexNet.The character image can be directly used as the input of the network,avoiding the process of image feature extraction and data reconstruction and improving the generalization ability.(3)In this thesis,by analyzing and comparing the current mainstream methods of character segmentation,it is proposed that the image of license plate is first clustered and then projected.The problem of character breaking in traditional methods such as projection or template matching is improved and the anti-interference ability of the system is increased.In this thesis,a complete vehicle identification system is designed and implemented,and the system is tested.The system includes modules of data acquisition,data preprocessing,system management,vehicle positioning,license plate positioning,character segmentation,character recognition and third-party interface call.