The Design And Implementation Of License Plate Recognition System

With the rapid development of both technology and economy, more and more people get their own cars. However, the increase of vehicles not only brings people convenience, but also causes troubles, such as traffic jams and accidents. Now, with the help of Intelligent Transportation System (ITS), we are able to achieve automated traffic management, and get effective control on the traffic environment. These will help to ease traffic congestion, ensure public security, and so on. Since License Plate Recognition (LPR) is the foundation of ITS, its importance is self-evident.The thesis describes the design and implementation process of a vehicle license plate recognition system. The topic comes from a project developed during the author’s internship. The author analyzed the system requirements, and completed the design of overall architecture of the system and the division of each module based on the requirements analysis. The system was mainly divided into four modules:The User Interface, License Plate Location module, Character Segmentation module and Character Recognition module. The User Interface was developed with MFC. Its main function is to get the message of the image that is supposed to be recognized, and show the result of the recognition to the user. The license plate location module got the binary image by dividing the source image into several parts, and getting each part’s binary image, and then it decided whether to use the Gaussian filter to process the source image based on each part’s binary image. After that, it connected hopping points to form connected area, and preliminary located the license plate from these areas. This method works well with backlighting, low contrast and fog problems. The character segmentation module used vertical projection together with template-based method to manage the segmentation of the characters. The module got the location of every character area by using projection method, and adjusted the location and the size of every area by using template-based method. The adjustment would be helpful to decrease the interference information for the character recognition. The character segmentation method achieved a high accuracy for character segmentation. The character recognition module used a Convolutional Neural Network (CNN) to recognize the character images. With a reasonable architecture for the Neural Network and enough training with good data for the training, the accuracy of recognition for every single character image is better than99%. The system described in the thesis is able to get a stable and accurate recognition of the license plate during the day and night in kinds of complex environments. The information it can get from an image includes the image of the vehicle, the image of the plate, the license plate number, the color of the plate, and the color of the vehicle.