Research On Vehicle License Plate Detection And Recognition

In recent years,with the rapid development of transportation systems,the develop-ment of intelligent transportation systems has attracted more and more attention.As an important part of it,license plate detection and license plate recognition will be helpful for the effective regulation of the traffic management department.At the same time,deep learning has achieved remarkable results in object detection and image recogni-tion.How to effectively improve the existing license plate detection and recognition methods using the deep learning technology,and improve the precision and speed of license plate detection and the accuracy of license plate recognition,this is of great significance for reducing the complexity of traffic management and building a more complete road monitoring and electronic police system.This dissertation mainly focuses on vehicle license plate detection and plate char-acter recognition.The specific research contents are listed as follows:(1)In order to improve the precision and speed of license plate detection,a license plate detection method based on the improved SSD model is proposed.The method adopts residual structure and multilayer perceptron to redesign the base network of SSD model,and adopts a model optimization strategy different from the original SSD model.The test and experimental results on the three subdatasets of the public dataset AOLP show that compared with the edge information based license plate detection method and the traditional CNN-based license plate detection method,the method respectively improves the average detection precision by 8.33%and 2.14%.Significantly,compared with the original SSD model based license plate detection method,the method improves the average detection precision by 0.45%and the detection speed by 2.54 times,and compresses the original SSD model parameter size to 43%.(2)In order to improve the accuracy of license plate character recognition,an end-to-end license plate character recognition method based on CRNN is proposed.The method is based on CNN and RNN.The convolution layer extracts the feature sequence of input image,the recurrent layer predicts the label distribution of each feature vector in the feature sequence and the transcription layer finds the highest probability label sequence and generates the final character recognition result of the entire license plate.The intermediate process of tilting license plate correction and license plate character segmentation in the traditional license plate recognition method is avoided,and the goal of end-to-end license plate character recognition is realized.The test and experimental results on the three subdatasets of the public dataset AOLP show that compared with the license plate character recognition method based on character segmentation and LDA classifier,the method improves the average character recognition accuracy by 4.33%;compared with the license plate character recognition method based on character seg-mentation and traditional CNN,the method improves the average character recognition accuracy by 1.62%,and the average license plate recognition accuracy by 6.05%.In summary,this dissertation makes an in-depth study on license plate detection and license plate character recognition.The proposed method significantly improves the precision and speed of the license plate detection and the accuracy of license plate recognition,and effectively reduces the parameter size of the license plate detection model.The purpose of accurate and fast license plate detection and end-to-end accurate recognition of character is achieved.