Design And Implementation Of Intelligent License Plate Recognition System Based On Convolutional Neural Networks

License plate recognition(LPR) system has assumed increasing importance in the field of modern transportation management and control system. Developing an LPR system which could deal with images taken from streets and internet is a challenging task because of complex background, various illumination condition and unfixed plate size. An LPR system usually consists of three modules: the plate location(PL), the character segmentation(CS) and the character recognition(CR). Studies on the problems of the three modules are carried on:For the PL module, to solve the problems of the edge extraction method, we propose the color edge method based on the color features of Chinese license plate and it improves the performance of the PL module. After that, we illustrate the significance of the parameters in the color edge method by means of experiments. Then, we introduce the morphology analysis to extract candidate regions from the edge image. With regard to the CS module, we propose the color depression gray conversion method to preprocess the input plate image. Then we apply the modified relocation method to get the accurate row and column information. After that, we implement the connected component analysis and the projection analysis. For the CR module, in order to solve the problems of the traditional convolutional neural network(CNN) model, we propose a simplified CNN and a recurrent CNN.Experiments show that the proposed LPR system achieves a PL rate of 98.95%, a CS rate of 96.58% and a CR rate of 98.09 which lead to an overall success rate of 93.74% and an average time cost of 318 milliseconds.