Research On Vehicle License Plate Recognition System

Vehicle license plate recognition system is one of important research subjects in the application field of intelligent transport, and is composed of license plate location, character segementation and character recognition. In this thesis, a technology solution was proposed on basis of analysis and study of vehicle license plate system's current research and development, in which approaches of license plate location based on color matches and texture, character segmentation based on image processing, and character recognition based on BP neural network were proposed. The application of these approaches effectively improved license plate system's real-time capability, veracity, adaptability and applied value, and extended applied extension.License plate location is the key technology of license plate recognition system. At present, license plate color and texture features were considered in the most license plate location methods, however, these methods had week adaptability in the different environment. In order to solve the problem, in the thesis, a new algorithm of license plate location was proposed, according to the fixed color matches and rich texture features of the vehicle license plate, and achieved accurate location of license plate. In the algorithm, a BP neural network was adopted to attain the strongly adaptive method which was used to recognize different colors in HIS color space, and for shortening the period of location, only the color space in the neighborhood of edge pixels were conversed, greatly shortening time cycle and improving location accuracy rate.In the character segmentation stage, in order to improve the validity, firstly, a series of image preprocessing algorithms were applied to color image. Considering character's color feature, character's density of effective pixels and character's aspect ratio, candidate character regions were demarcated. And then, according to the characteristic of characters' distribution in the different license plate and the features of characters, an approach of level histogram projection was successfully used to distinguish single/double row(s) characters license plate, making the adaptive of character segment stronger.Character recognition, which affects the entire performance, is the last stage of license plate recognition. In the thesis, firstly, according to the characteristic of characters' distribution in the license plate, these characters would be divided into four styles, and the method of scanning by pixels were adapted to extracting features of classified character image block. Then, in accordance with different types of characters, different BP neural networks structures were created, and successfully achieved high accuracy rate of recognition.The experiments showed that the accurate recognition of the different types of licese plate could be achieved quickly in the complicated environment and different illumination, making use of the proposed technology solution about vehicle license plate recognition system in the thesis and having value on application and popularity.