Study And Improvement Of License Plate Recognition System

With the continuous progress of society, sharp rise in the number of cars in China has brought about enormous pressure to the traffic management of city road, and the previous manual management mode is time-consuming, inefficient and has a greater probability of error identification, making intelligent transportation develop rapidly. License Plate Recognition(LPR) technology has been an core subject of intelligent transportation, therefore the study of the license plate recognition technology is of great significance.There are a series of problem in license plate recognition under the weather conditions of rain and fog, including vague image, obscure character boundary,and change of light, which brings about a mount of difficulties. In order to solve the problem, study is made on the license plate recognition system in this condition. As the key technology of intelligent transportation in the future, the entire process of license plate recognition includes three parts of license plate positioning, license plate segmentation and license plate recognition. The mathematical model is build and the simulation experiment is carried out by detailed analysis of entire procedure of license plate recognition including positioning, segmentation and recognition. License plate positioning method based on color is emphatically expounded, to which the Gamma correction algorithm and the denoising algorithm of color images are added, making the algorithm more practical. Traditional character segmentation methods and binarization methods are introduced, and a kind of binarization method in double color models is proposed in this article according to the feature of characters. Segmentation method based on connected area detection is adopted by this paper to achieve accurate character segmentation.The final license plate recognition accuracy is determined by the part of recognition. In recent years, with the continuous development of the core algorithms, especially the rapid development of intelligent algorithms, license plate recognition accuracy and recognition speed is further improved. In this paper, study on license plate recognition based on multiple traditional neural network algorithms and corresponding plate recognition system is realized by this article. The modal structure and data processing performance of various kinds of neural networks are analyzed, and an adaptive fusion algorithm based on BP, RBF and GRNN neural networks is proposed for the recognition of license plate characters in the paper. Combined with the advantages of multiple neural networks, this algorithm can avoid some situations that a single neural network fail to identify characters or has identification errors, thus further improve the accuracy of license plate character recognition and provide a reference for the follow-up study on license plate recognition.