Research On License Plate Recognition Technology Based On Radial Basis Function Neural Network

Intelligent transportation system is a cutting-edge research subject in the current field of transportation, and the license plate recognition is one of the core technologies to realize intelligent traffic management, which is of important application value in intelligent transportation systems. License plate recognition system mainly includes license plate location, character segmentation and character recognition. Neural network recognition is an important research direction in the field of pattern recognition, which has the features of self-organized learning, high fault tolerance and Strong robustness, then it’s Very suitable for license plate recognition. The main job of this paper is that applying radial basis function neural network (RBFNN) to license plate recognition systems to improve recognition accuracy and recognition of license plate recognition system speed. The main job is as follows:Firstly, a comparative study is done at license plate location method based color images and gray-scale images, and the paper adopts the license plate location method based on Adaboost algorithm con siding of the disadvantage that color images are affected easily by the light and it’s has a longer positioning time. A plate has been grayed before it’s positioning, and three commonly used Haar features are expanded in the process of positioning in accordance with the characteristics and texture of the license plate in our country in the process of positioning, Defines a new four rectangular Haar condition characterized. Experimental results show that the method not only shortens positioning time, but also improves the plate positioning accuracy.Secondly, dividing the license plates which have been positioned, and finalization processing has been done before the divide in order to extract the image information facilitate, then process the license plates with tilt correction and median filtering. As the disadvantage of methods by vertical projection template matching which are commonly used, a new dividing method of combining the vertical projection and multi-template matching is proposed, The method is accurate character segmentation rate of98.35%, it is higher than the vertical projection segmentation method and template matching segmentation method.Then Hog feature extraction has been done at the characters which have been divided well, then I make a brief introduction about the common used methods of the license plate recognition, and a comparative analysis is done at the advantage and disadvantage of BP neural network and RBF neural network, Finally adopted RBF neural network to identify the license plate characters and use genetic algorithm weights RBF neural network optimized for the purpose of improving approximation of nonlinear network capacity to ensure recognition accuracy and recognition.Finally, the simulation of the network proves that it’s of strong network stability optimized and high recognition accuracy, and the license plate recognition system is completed finally.