Research On Recognition Method Of License Plate Via AdaBoost And BP Neural Network

With the development of urban traffic,vehicles are growing continuously.Vehicle license plate recognition technology(VLPR)is the foundation of the intelligent traffic system which is the one of main direction in the construction of smart city in the future.This technology can be applied for transportation scheduling management,vehicle automation management and monitoring etc.It will have positive influence on reducing the traffic management of artificial cost,improving the efficiency of the society etc.In all kinds of license plate character recognition methods,the BP neural network is widely used for its good robustness,fault tolerance and adaptive learning ability.At the same time,it has the ability of processing and distributing information in parallel.However,it has limitation.This thesis studies BP neural network method and gives some suggestions to improve character recognition methods.The following is the main research and achievement.(1)In this thesis,the hierarchical approach is adapted to locate license plate step by step,by continually narrowing target region to increase the accuracy locating.Firstly,it uses License plate texture features to check and locate license plate roughly.Secondly,Hough transform is adapted to adjust those license plates which may be tilted.Generally,roughly locating can check the complete plate outline,and Hough transform usually checks straight line which is on the edge of plate.Finally,author get data for License plate area without border by analyzing the plate background color with the HSV color model and adapting FloodFill algorithm to locate license plate accurately.(2)Firstly,based on analysis of a variety of image binarization segmentation method,and combined with the feature of the color of the license plate,this thesis puts forward a new method which using color similarity clustering on binarization segmentation method for license plate image.Secondly,by analyzing common license plate segmentation algorithm and adapting Vertical projection method,license characters are segmented.Finally,the centre-of-gravity method is used to normalize the position of character images and bilinear interpolation is used to normalize the size of the characters.(3)In this thesis,several common character feature extraction methods are introduced,and the advantages and disadvantages of each method are analyzed.A New method of feature extraction is put forwarded by the experimental analysis.In terms of character recognition,through analysis of advantages and disadvantages of the BP neural network,and as to the BP neural network is easy to fall into local minimum and has slow convergence speed and many other problems,the research puts forward that it can improve the level of recognition of character by applying the AdaBoost algorithm into the BP neural network.The experiment results show that the proposed BP neural network based on AdaBoost enhancement can overcome the insufficiency of the neural network classifier and effectively improve recognition rate of the license plate character.