Arithmetic Research And Realization Of Vehicle License Plate Location And Segmentation

Vehicles' License Plate Recognition System (LPRS), which is an important part of the contemporary Intelligent Transportation System (ITS), can be applied to vehicle management situations of all levels and all kinds. Compared with traditional vehicles managements, LPRS has greatly improved the efficiency and level of management and saved manpower and material resources, laying a good foundation for the realization of standardized management. We can safely come to the conclusion that LPRS has already improved the order of the traffic system, illustrating a good prospect of application for us. Generally, the LPR system consists of three modules: license plate location character segmentation and character recognition. Its study concerns various disciplines including Pattern Recognition Artificial Intelligence Computer Vision Digital Image Processing and so on. It is the location and segmentation of license plates standing at the heart of LPR system. Considering that the complexity of image background and the uncertainty of plate position and image quality, it is necessary to do further research into it.By summarizing the latest research achievements and development in the area of license plate location and segmentation both here and abroad, this paper, after making a deep comparison between the intrinsic characteristics of license plate and the current location and segmentation technologies on it, proposes its own understanding and designs a new LP location and segmentation system.In allusion to the deficiency of the single model location, a new positioning method concerning the license plate, which is based on multi-color models, has been presented in this paper. Firstly, the input image of RGB is transformed into HSV and YIQ color spaces, from which comprehensive information will be obtained. Further work can be done: by integrating this information, image segmentation will be able to move out large numbers of disturbing background message, leaving those smaller areas (license plate itself is included) whose colors are similar to it, paving the way for license plate location. After that, license plate can be correctly located through the "two-step-location" method. In the first step, the area to be researched will be reduced greatly by the help of section thinking which makes coarse positioning come true. In the second step, the rough located images can be divided into two situations: those in which the color of the target areas of a license plate is quite different from the non-target area and those in which the color of the target area is alike to the non-target. In order to locate the license plate precisely, the method of color segmentation and that of edge detection based on Log operator are adopted respectively.Considering that many factors varied from bad weather, diversified illumination to shooting angles might cause series of problems such as stains, slant and so on—it might be very easy to make mistake when segmenting character directly- this paper proposes a preconditioning algorithm based on character segmentation to take precautions against it. Equipped with those collected sub-images, techniques of winner filter and histogram equilibration have been adopted to solve stains and illumination problems. To the case of slant, the Hough Transformation and technique of projection will been employed, followed by three further operations: removing borders, binarization and unifying the background color of license plate, all of these operations would lay a good foundation for character segmentation.Since the proportion of the size of a license plate is quite constant, when it conies to the work of character segmentation, a technique of character segmentation based on the characteristic of the proportion of the license plate has been proposed in this paper. Obviously, we can take advantages of both the vertical projection coming from binarization pictures and the proportion characteristics of the plate itself to segment characters. First, roughly ascertain the intersected areas between characters according to its proportion characteristic. Then find out the best segmentation point—the locality with the smallest vertical projection numerical value. This method can perfectly solve the problem of cohering characters.In this study, 109 vehicle images of yellow background and 344 ones of blue background have been selected and put into use in the location test. Among the former ones, 104 images have been tested correctly by the technique of rough location, following that 100 ones by the technique of secondary color segmentation and 98 ones by Borderline-Measurement Technique based on log operator. In comparison to it, 311 images of the latter ones have been tested correctly by the technique of rough location, 303 ones by the technique of secondary color segmentation, and 306 ones by Borderline-Measurement Technique. In the rectification stage, 223 images of license plate inclined horizontally have been located and then put into use in the test, with the ratio up to 96.9 %. At the same time, 118 ones inclined vertically have been oriented carefully, later on, these images have been rectified vertically, and the ratio has achieved 91.5%. Then, after image preprocessing and the last act of fine orientation, 170 pictures have been obtained and participated in the test of character segmentation and the correctness ratio has reached 95.3%. The experimental results indicate that the technique of license plate positioning and character segmentation proposed in this paper is good enough to meet the practical requirements...