Research And Realization Of License Plate Recognition System

With the rapid development of social economy, roles of the Intelligent Transportation Systems (ITS) are getting more and more attention in urban traffic management. As one core component of ITS, License Plate Recognition System (LPR) has a very broad application prospects in many aspects, such as real-time traffic monitoring, highway automatic charging management system, traffic security monitoring and so on. At the present, LPR is focused at home and abroad.As many unfavorable factors such as the outdoor all-weather working conditions, the complex and variable plate itself etc., LPR is one of the most difficult problems in the field of pattern recognition for a long time. In this thesis, based on the in-depth study to the LPR, one set of effective solutions is proposed, and an actual software system is developed. The research work and achievements of this thesis are as follows:(1) To process the initial input image, a variety of image preprocessing techniques is used and an adaptive binarization algorithm based on the mean of each line in the image is proposed. Through this algorithm, the region which have character can be effectively highlight and a lot of noise can be filted at the same time. So the algorithms including the jumping point detection and connected domain filtering algorithms can be easily used to locate the plate region. In addition, a comprehensive evaluation standard combined with subsequent processing steps is proposed on the basis of many experiments, which can effectively find the real license plate.(2)To effectively solve the tilt correction and precise horizontal location of the horizontal position for the plate in the LPR, an algorithm which is based on edge detection and improved K-means clustering is proposed. The tilt angle of the license plate and the horizontal position of the character in the tilt corrected image can be find at the same time through this algorithm. In addition, in order to get good effect in the license plate image binarization, an improved binarization algorithm based on the OTSU is proposed, through which the fault of the OTSU binarization algorithm can be effectively avoided when binarizing the plate image. The effect of this algorithm is satisfactory in the experiments.(3) To get the character from the license plate, An algorithm which combines the vertical projection, connected components detection and license plate template is proposed, what is more, an evaluation function used for the template matching is proposed. The noise impact and the error caused by the adhesion and fracture of the character can be effectively avoided throught this algorithm.(4) To achieve the purpose of feature extraction and data reduction, the wavelet transform is used to deal with the character image. In addition, a character recognition system using multiple neural network is designed, which can effectively classify the characters on the license plate.(5) In order to test the performance of the algorithm, a more complete software for the license plate recognition is developed in the PC-platform.In this paper, a lot of work of the divisions and classifications is did based on the current license plate database. The effect in the experiments shows that the algorithms proposed in this paper are advanced, effectively, and the software designed is practicality. In addition, the full analysis and comparison to the algorithm in this paper combined with previous results achieved is made. The results show that the system designed in this thesis has good indexes of accuracy and real-time. It has good robustness and adaptability for fuzzy images. The average running time is about350ms. And the recognition rate is close to89%., which has reached the level of engineering application.