节点文献
基于智能信息处理方法的车牌识别算法的研究
Study on License Plate Recognition Arithmetic Based on Intelligent Information Processing Technologies
【作者】 李垠;
【导师】 赵继印;
【作者基本信息】 吉林大学 , 通信与信息系统, 2008, 硕士
【摘要】 车牌识别系统LPR是智能交通系统的重要组成部分,主要应用在车辆自动化管理领域,包括高速公路的不停车收费、停车场管理、多用途收费系统以及各种违章车辆监控场所。由于图像背景的复杂性以及图像质量的不确定性,LPR系统仍有许多技术难点未能得到很好的解决。本文针对车牌识别系统中的车牌定位和字符识别两方面技术问题进行了系统的研究。综合运用数字图像处理技术,提出了基于投影法和遗传算法相结合的车牌定位方法。首先,利用投影法定位出车牌的左右边界,再利用设计出的两种模板滤除点状噪声和线状噪声,最后以二阶矩作为目标函数来定量描述车牌区的纹理,使用遗传算法搜索出车牌区域。研究了人工神经网络和BP算法的基本原理,从车牌本身特点出发,设计出分别用于识别汉字字符、字母字符、字母和数字字符和数字字符的四个子网,对字符的输出进行二进制编码,然后,针对几个结构相似的字符专门设计了小型网络以实现对它们的精确识别。试验结果表明,本文提出的车牌定位方法定位准确率高,抗干扰能力强,对于质量较差的图像也有较好的定位效果。本文设计的BP网络分类器具有识别率高,稳定性好,容错能力强等优点。
【Abstract】 1.IntroductionNowadays, the fast development of transportation has been greatly improving our live, but the traffic jam, the traffic accident and entironment pollution become more and more serious and the ITS is one of the schemes which resolve these problems. Vehicle plate recognition technology is the key technology of ITS, it is also one of the important research subjects in ITS department. It can supervise vehicle automatically, memorize the vehicle plate and recognise the vehicle plate characters. It has expansive apply foreground in freeway automatism charge, residential area safety and electron policeman.Plate location and vehicle plate characters recognition are key technology of Vehicle plate recognition technology, the internal scholars and overseas scholars have done much reseach about it, but the effect is not very good because of the condition of beam, the standard of vehicle plate and the quality of image. The same production is not appropriate between different area. Especially in our country, the standard of vehicle plate is not uniform, the conditon of road is complex and the circumstance is always mutative, so developing an all-purpose and accurate vehicle plate recognition system is very difficile. At present, the“Eye of HanWang”of HangWang company is full-blown production. In addition, Asia Vision Technology Company, Jitong Electron Technology Company and Zhongzhi Traffic Electron Technology Company all have their production. Other scientific research institution also develop this reseach, including the Image Processing and Recognition of Xi’An JiaoTong University, the Computer Science & Engineering Department of ShangHai JiaoTong University, Tsinghua University and ZheJiang University.Genetic Algorithm and Artificial Neural Network are parts of Intelligent Intormation Processing Technologies. Genetic Algorithm is high efficient, parallel, adaptive global optimization probability searching method. It provides an universal scheme which is solved complicated optimization and it produces strength robust. License plate location is difficult because of the beam and background, then, using Genetic Algorithm can well solve this problme. Artificial Neural Network has the ability of parallel processing, distributed infor- mation memory, self-adaptation, self-organization, associative learning and admitting-error. It can well solve the difficult license plate location’problem induced by beam and background. In a word, Intelligent Intormation Processing Technologies provide good method for uncertain, fuzzy, complicated information processing problem such as vehicle plate characters recognition system.2.Research ContentThis paper has studied a lot of method and theory about vehicle plate characters recognition system , considering the characteristic of our vehicle plate, using Intelligent Intormation Processing Technologies to solve difficulty in vehicle plate characters recognition system. A suit of scheme about vehicle plate characters recognition system was proposed. Main research contents are as follows:(1) Analyzing about the characteristic of our vehicle plate. There are three kinds of objects about vehicle plate characters, there are Chinese characters, letters and number. And this paper only discuss the civil vehicle plate. The Chinese characters of civil vehicle plate contain about“京、津、晋、冀、蒙、辽、吉、黑、沪、苏、浙、皖、闽、赣、鲁、豫、鄂、湘、粤、桂、琼、川、贵、云、藏、陕、甘、青、宁、新、渝”(31), the number contain about“0~9”, the letters contain about from“A”to“Z”, besides“I”and“O”. Considering about this condition, four kinds of sorters is designed to recognise the characters.(2) The method of image processing is introduced and realized. This paper introduced the method of graying, binaring, filtering and edge detection. Using Sobel operator to detect edge. The Sobel operator is conformable for license plate location and skew correction.(3) After discussing the theory of Genetic Algorithm, the method of license plate location base on projection method and Genetic Algorithm was proposed. First of all, the edge of the image is detected, second, the image is binard, so that small texture noise could be eliminated. Using projection method to locate the left and right borderline of vehicle plate. Because of the symmetry of the image of vehicle plate, whether what is beam and condition like, the projection map of the image is approximate symmetry. Because of this reason, using projection method to locate the left and right borderline of vehicle plate is very easy. After locating the left and right borderline, the image is single. The texture of the vehicle plate is more clear than other area. At this time, use the mask filter to process the image, then the point noise and line noise will be eliminated. The vehicle plate can be found by Genetic Algorithm when individual and generation are less. Experimental result indicates that this method is very effective and accurate,the location accuracie is 97%.(4) Skew corrected, character segmentation and normalization are very important before character recognition. Before skew correction it is essential to enhanced the contrast through histogram equalization,so the edge of the image will not disappear after Hough transform. And after that, use Hough transform to correct angle of the image. Use scan method to eliminate the frame and rivet of the image, so the character recognition will be easy. Use bilinear to realize normalization of charater size. The location of charater is normalised base on external frame of chareter. piecewise threshold binaring method was proposed. First, calculate the average, max and min of pixels, second,if the average is smaller than 110, add max and min together and divided by 2,then the value is the threshold, otherwise, the average is the threshold. Experimental result indicates that this method is very effective(5) After discussing the theory of BP neural network, designed BP neural network Classifier for vehicle plate characters recognition system. According to the Characteristics of vehicle plate characters, designed four kinds of BP neural network Classifiers to recognise the chinese characters, letter, letter and number, number separately.And every character has been coded. At same time, especially designed small BP neural network for 8、B、D、Q、0 which are promiscuous and this method is effective. The character image is eigenvector of BP neural network. Experimental result indicates that the character image is the best character, different size strokes, different font, different character location and fracture character will not influence the precision of recognition, more interference can be avoided.3.ConclusionLicense plate location and vehicle plate characters recognition are key technology of vehicle plate characters recognition system. This paper is mainly about them. The characteristic of projection method are easy and fast. The characteristic of Genetic Algorithm are high efficient, parallel and adaptive. Using Genetic Algorithm can find the optimal solution qucikly. Combining these two methods can take full advantage of their strongpoints. The location accuracie is 97%. Artificial neural network is especial for pattern recognition. According to the characteristic of vehicle plate characters, designed four kinds of BP network to recognise the different kinds of charater. Especially designed appropriative BP network for those easily promiscuous characters and recognition precision has been improved. The recognition accuracie is 96.5%.
- 【网络出版投稿人】 吉林大学 【网络出版年期】2008年 10期
- 【分类号】TP391.41
- 【被引频次】4
- 【下载频次】395