Study On Vehicle License Plate Recognition Method

Intelligent Transportation System (ITS) has been developed for 20 years, which has been paid more attentions in many countries. License plate recognition (LPR) is the one of most important parts of ITS. Nowadays, LPR system has been widely used in highway monitoring, management and entrance fee. and so forth. In this study, a license plate recognition system is designed and the related implementation technology is further studied.At first, the license plate recognition technology and its related methods in literatures are briefly introduced, in which these approaches have been given a series of comparative analysis. This study proposes some improved methods for license plate location, binary method by using dynamic sliding windows (DSW), degraded character restoration and the basis character feature recognition method.In license plate location part, the images obtained in actual traffic intersection, is handled with the Sobel operator, line scan and fast searching method to complete a rough positioning. A HSV color space combining with plate aspect ratio has been proposed to realize a precise plate positioning. The method makes full use of the license plate texture, geometry and physical characteristics, and removes the former positioning of plate. The experience results show that the proposed method presents fast and accurate performance for actual images in natural light and other conditions.In license plate processing part, the most important step is to obtain the binary images of license plate. For this purpose, this article proposes a novel adaptive binary technique named DSW based on the existing Otsu method. Through a posteriori evaluation of the binary result to estimate the dvnamic frame size, this method overcomes the flaw of Otsu method with one threshold in whole image in order to describe the "local" irregularity in the plate image such as shadow and uneven illumination.In plate characters processing part, a vertical projection for character segmentation has been used, which applies a connected components to detect fragments, removes noise and merges fragments that belong to the same character.In character recognition part, this article presents an algorithm based on LSSVM for the multistage classification recognition. The characters are separated into three categories by their characteristics, and a grid feature vector extraction method is proposed. The experimental results show that the proposed method has better performance and is effective for the ambiguous characters, less time consuming and the downsize of training dataset.