Research On The Application Of Genetic Algorithm And Neural Networks In License Plate Recognition System

With the development of the city Intelligent Transportation Systems (ITS) in our country these years, License Plate Recognition (LPR), one of the information collecting means in city traffic management systems, has been rapid development and at the same time there are still some shortcomings.In this paper, a comprehensive discussion of several key technologies of the LPR is expounded firstly. And then, combined vehicle license location technique and character recognition with genetic algorithm and neural network, and based on the license plate recognition system at home and abroad, a new License Plate Recognition System is designed for the license plate characteristics of our country. The system can effectively improve the accuracy of license plate location and the efficiency of identification computing.This thesis includes the following three aspects:(1) According to gray-scale and color characteristics of the vehicle license plate region, improved location algorithms are proposed based on gray-scale and mathematical morphology, and based on color characteristics respectively. Genetic neural network algorithm is applied to the plate location areas by constructing the fitness function rationally.(2) Aimed at the shortages in license plate location and tilt correction of current mainstream algorithms, a more effective improved algorithm using Hough transform method is proposed in order to confirm the tilt angle, and then is compared with the former algorithm.(3) Advantages and disadvantages of template matching and neural network methods are analyzed and compared. Combined template matching with BP (Back Propagation) neural network, three-layer Feed-Forward Neural Networks (FNN) of numbers, letters, numbers mixed letters and the Chinese characters classifier is designed. The improved BP algorithm increases the convergent speed and greatly reduces the network training time. The experiment results show that this method is effective to the similar character recognition and accuracy has also been greatly improved.