Research On Real-Time License Plates Segmentation And Recognition Techniques

Auto Vehicle License Plate Recognition (LPR) is one of the key component parts of Intelligent Transportation Systems (ITS). LPR recognizes the vehicle images acquired by the high-definition cameras automatically, without affecting the order of the driving cars. Also, it can reduce the traffic accident frequency and decrease the laying off of personnel or the elimination of resources, managing the traffic effectively. Based on the computer vision method, the research of LPR system is an untouched inspection method which crosses image processing, pattern recognition and artificial intelligence synthetically. It has been widely used in the detection of drive against traffic regulations, highway charge management, and intelligent parking management. The LPR system also helps to improve the control capability to automobile and it is one of the important contents for realizing automobile intellectualization.The efficiency and real-time property of LPR system not only concern to electron optics, imaging, detection, communication, machinery, electronics, but also involve image processing. It is an integrated system of hardware and software. Typically, the LPR system composes of three main modules: license plate location, license plate segmentation and char recognition. Based on the analysis and research home and abroad, in accordance with the concrete characteristics of Taiwan license plate, this thesis surveys the license plate segmentation and recognition problems under the complex enviroment. Following is the specific content.1) Firstly, this thesis gives analysis to the research background and research significance. The development of LPR system at home and abroad is reviewed and the general technique is summarized in detail. The research significance is also given in this thesis.2) The license plate detection is the basis of follow-up processing. Hence, a brief review of license plate detection technology is given in chapter 2. A succinct introduction of the general license plate detection is also given. We mainly introduce two method of license plate detection. We summarize the general detection technology briefly.3) Character segmentation can be dividing into three parts which are license plate binarization, revise the image tilt and segment the characters. In the paper we focus on the segmentation task of license plate in Taiwan Area of China. This paper proposes a local binary algorithm which counters the requirement of real time and precise of the system. Comparing with other methods, experiment test shows that proposed method has been improved the binarized quality on the basis of necessary processing time consumption. The comparison results of other methods are also list in the table.4) This thesis proposes two resolution of slant correction. One method is based on the piecewise linear fitting and the other one is based on the transformation of Radon transformation. The former method relies on the quality of binarization and the latter doesn't. Each method has both advantages and disadvantages. The piecewise linear fitting method adjusts to the small angle range while the Radon method adjusts to the large angle range.5) We propose a fast and quick character segmentation algorithm for the characteristic of license plate in Taiwan Area of China. The char image width is estimate by the height of the image to be segment firstly. We search the character edge based on the vertical projection feature. The accurate char width is modified by the secondary segmentation. Proposed segmentation is fast and simple in computation, which can server as references for the other segmentation method.6) There may be fake char images and real char images which exist simultaneously in the splited images. A method based on the structural features of binary images and the color information of the color images is proposed to distinguish the real images from the mixed images of license plate char images. Combining the extracted structural features and the inherent characters of vehicle plates, Most of the fake char images are removed. At last, the color information of the fake images which are difficult to exclude from structural features will be analysis in HSV color model, and the color images which exist huge differences from others will be removed at last. We get the real char images by the prior nudge of margin flag which exist in the license plate.7) As to the aspect of license plate recognition, this thesis proposes a cascade classifier based on SVM. And the experiment result shows that the proposed algorithm can effectively improve the recognition rate of confusing characters with nice recognition rate and high performance.In the last part of this paper, major achievable outcomes by researching are summarized and a few solved most issues are advanced in farther research for the future with advantages and disadvantage summarized.