| Vehicle License Plate Recognition System (LPRS) using computer do processing, analysis and recognition for video traffic images to extract plate information. Objective is to facilitate management such as traffic control, statistics, and fees. And LPRS is one of the hot core issues of intelligent traffic management. Many scholars at home and abroad are specializing in license plate localization and character recognition technologies, and have achieved many results, but there is still much to continue to study and improve. Such as most of the license plate recognition methods have a high recognition rate for the high-resolution images, but can not solve the all-weather problems, poor availability and low accuracy. So it is meaningful for us to improve the robustness of the LPRS for all-weather images, and is worthy of further study.License plate recognition technology mainly consists of three links, license plate localization, character segmentation and character recognition, and the three links are carried out in turn. The topic is hammering at the technologies of the three links in view of the characteristics of all-weather video traffic images. Firstly, in part of the license plate localization, in order to improve the locating speed, remove the processing of enhancing image quality. The topic use canny operator to test edge which can detect weak edges for low-resolution images, but use sobel operator which is easier than canny operator for high-resolution images in order to avoid too much interference, and sobel operator can smooth the noise. After dynamic binarization, this topic presents a method which can remove noise through line scan mark and mathematical morphology. This method fully uses the license plate texture characteristics, and can quickly locate candidate regions of license plate. The topic also presents a corner detecting method which can get location information of candidate regions. The points which need to calculate in this method are very little. This method greatly reduces the time compared to the traditional mark method of connected regions. In part of the fine locating, because there is geometric correction processing before this part, the cutting-edge method in this topic has a high positioning accuracy which using texture and vertical projection. Secondly, to solve the problem of adhesion and fracture of characters, this topic presents a vertical projection method based on fuzzy decision. The method can add or delete split points automatically, and can solve the problem of non-connected Chinese characters and touching characters. Thirdly, this topic extracts eight feature vectors of 33 numbers and letters to build feature set, and completes character recognition by similarity of feature vectors, and then uses feature point matching method as the second recognition of similar characters. Character feature vector in this topic has a certain tolerance for the geometric expansion of the character image. For the stroke adhesion problem of Chinese character, this topic presents a Chinese character recognition method based on fuzzy outline. This method extracts the fuzzy outline of the Chinese character firstly. And then calculates the Fourier descriptors of the fuzzy outline. This method does not rely on the internal character stroke information, and extracted Fourier descriptors have invariance on rotation, translation and scale, so it has good robustness. This topic carries out the algorithm programming in Visual C++6.0. The experimental results show the license plate recognition method in this topic has a good robustness for all-weather video traffic images, and meet the requirements of real-time. Compared with the traditional identification methods, the overall recognition rate has greatly improved.This thesis consists of three parts. The first part is about the background and significance of the topic; the second part is the introduction of the methods which are used and improved in the topic; the last part is the simulation results and prospects for future research. |
PDF Full Text Request