摘要
汽车牌照定位与识别系统是智能交通系统的一个重要组成部分,利用该技术可以实现对车辆的自动登记、验证、监视和报警,高速公路收费,对停车场进行管理,特殊场所车辆的出入许可等。
本文主要讨论软件部分的工作,包括汽车牌照定位分割、字符切分和字符识别。首先通过对图像预处理的各种方法及特点进行了研究,找出了适合本系统的汽车牌照图像的预处理方法。其次对汽车牌照定位分割的主要方法进行详细地介绍和比较,确定了汽车牌照的定位分割过程。先采用线扫描法对汽车牌照进行粗定位,然后在图像处理的基础上采用基于边缘检测和投影图法对汽车牌照完成精确定位。再次实现汽车牌照字符的切分,为保证切分效果,对定位后汽车牌照进行倾斜校正,然后采用垂直投影法实现字符切分,实现字符切分后对字符归一化处理,为字符识别提供大小一致的字符。最后本文采用结构和统计相结合的方法对汽车牌照字符进行一种基于模糊决策的模糊分类器的模糊决策识别。在此理论研究的基础上利用VC++6.0开发了汽车牌照定位与识别系统的软件。
实验数据表明,本系统能较准确地定位分割车牌、字符切分和字符识别。同时也说明了把多种图像的预处理与识别技术有机结合起来可以提高系统的识别能力,在有效、实用原则的基础上利用模糊技术不仅可以提高汽车车牌上汉字、英文字母和数字的识别率。
该系统的开发适合我国目前交通汽车业的现状,具有很好的市场开发价值。同时该系统涉及到数字图像处理、模式识别和模糊识别等多个技术领域。因此同样具有很高的理论研究价值。
Along with our country reform and open policy unceasingly thorough, the national economy develops fast, the automobile popularization is getting higher day by day, with the traffic flow increasing, such phenomena as traffic jam, traffic accident, and air pollution are getting worsen correspondingly rapidly. Carrying out intelligence transportation management is imperative in China. The license plate localization and recognition systems are important components of intelligence transportation system that can make it possible to manage the vehicles automatic log-on, the confirmation, the surveillance and the warning, the highway charge, the parking lot, special place vehicles permission and so on.
The license plate localization and recognition system research involves the digital image processing, the computer vision, the pattern recognition and the artificial intelligence and many other technology areas, its key technologies include car license localization division, character segmentation, and character recognition. The license plate localization division refers to the location of license plate area from the primitive gathering automobile image and division, is foundation of the license plate recognition. The character segmentation duty is cutting many lines or multi-character images into the single character from the entire image, further carries out the character recognition. The character recognition is a process of understanding Chinese character, the letter and the digit in the license plate, is core in the system.
License plate localization division, character segmentation and character recognition has been deeply researched in the paper, the design and software flow based on image processing, pattern recognition, fuzzy recognition of license plate localizations and recognition system has been proposed. Application of image processing technology, the pattern recognition technology and fuzzy recognition mainly has been discussed in the license plate localization and recognition system, on the basis of the primitive automobile image pretreatment technology the license plate localization division and character segmentation has also been researched deeply and the pattern recognition and the fuzzy recognition technology has been used to carry on the recognition to license plate single character.
This system uses the gradation image processing to finish the localization and the recognition. Therefore, first the system transforms the color image the gradation image with the weighting even averaging method. In order to reduce the background picture element the disturbance, but preserves or strengthens the picture element in the target area ,the image should be carried on binarization processing, this article uses the quite common overall situation threshold value criterion, when the system carries on the threshold value choice, it adopts the mean value means, for the prominent license plate edge, it uses the gradient value improvement algorithm to realize the image gradient peaking; the middle value filter remove the image noise and the disturbance factor, meanwhile it has realized image smooth.
License plate localization division is the first step of this system processing, whether the localization is accurate or not is directly relation to success or failure of the system recognition. According to advantage and shortcoming of the present existing localization division methods and reference of human eye recognition license plate process, namely found the approximate position which the first license plate is, then further determine again the license plate concrete position, and further determine license plate localization division process.
This article mainly introduced our country license plate essential feature, the localization knowledge and the main localization method to license plate localization division part. Elaborated the license plate localization method which this article realizes with emphasis, first uses the line scanning method to carry on the thick localization to the license plate, then to the histogram Gauss threshold value law, the overall situation dynamic threshold value law two division methods carries on the comparative analysis, selects the overall situation dynamic threshold value law that suited this article to realize the division, finally uses the methods based on the marginal check and the axonometry to complete the pinpointing to the license plate. This method can adapt different complex background and the different environment license plate zone location, it has the good robustness and timeliness, the localization is accurate.
License plate character segmentation directly effects character recognition, firstly the correction for grade pretreatment of located license plate should be finished before license plate character segmentation. The vertical projection has been used to realize the character segmentation, this method can solve effectively major problems that can affect such factors that character segmentation is not entire in the algorithm, and segmentation is shifting. In order to reduce the single license plate character size variables which may bring difficulty for the following license plate character recognition, car license character size normalization processing is finished by pair of three transect interpolation, the normalized effect is better.
License plate character recognition was most important link of the overall system, the recognition methods quality immediately influence overall recognition result. This paper uses characteristic selection based on the statistics and the method which unifies characteristic selection based on character structure to finish the recognition to the license plate character, the methods firstly withdraws the four complex index of stroke, area, complexity, bland ratio, these seven character characteristics to the license plate character, to reduce as far as possible distorts to the information characteristic disturbance, or withdraws the reliable characteristic information from the distortion character, defines 5 basic elements that constitutes character outline by using outline first-order differential change tendency, and gives the element examination rule, it may also be insufficient to use the structure element to distinguish incomplete digit accurately, outline statistical nature has been introduced. By designing fuzzy classifier based on the fuzzy decision-making according to above characteristics, the classifier can finish recognition to the single character.
Finally, in the above work achievement foundation, the license plate localization and the recognition system software that can collect license plate image pretreatment, the localization division, the character segmentation and the character recognition function has developed by using VC++6.0 .This software has such characteristics as modulation, visualization, simplification.
This system makes use of the quite clear ordinary license plate image to carry out the localization, the segmentation and the recognition and make quiet good progress; there is no the scene experimental condition, therefore timeliness and the usability of this method should further be studied through the scene experiment. Moreover the localization and the recognition is based on the gradation image, does not make use of the license plate original color characteristic, this may affect system performance to a certain extent. The color image contains the more pictorial information compared to the gradation image, the research of the localization, the character segmentation and the recognition based on the color image should be strengthen to enhance the system performance.
At present, the license plate recognition technology is at the start stage in our country, although the recognition success ratio of each stage arrived at above 90%, but could not have achieved practical request that Synthesis recognition rate can reach ( recognition rate 96% and above the recognition speed of 3 cars per second).
The system can locate, divide car license accurately and carry on the recognition. The system suits our country present transportation automobile industry situation, it can widely be applied in such places as highway, urban road, parking lot, it is helpful to our country traffic control automation advancement. Therefore the system has very good value in fundamental research and market development.
引文
[1]段里仁.智能交通系统在我国道路交通管理中的应用[J].北京智能交通系统发展趋势国际学术研讨会论文集.97:15-22.
[2]刘允才.智能交通国际发展概况和国内优先考虑的课题[J].公路,2001,11(11):26-34.
[3] Liu Jilin,Ma Hongqing,A High Performance License Plate Recognition System Based on the Web Technique[D].
[4]林立,何为,韩立群.汽车牌照自动识别技术的现状与发展[J].北京轻工业学院学报,2001 , 1 9(1):36-38.
[5]王广宇.汽车牌照识别系统综述[J].郑州轻工业学院学报(自然科学报) ,2001.16(2):47 -50.
[6]高一凡,王选仓.条形码在交通管理中的应用[J].交通与计算机,1996 ,14 (2)53 -58.
[7]赵明生,包丛笑.车牌识别技术在交通工程和城市安全中的作用及应用前景[J].交通系统发展趋势国际学术研讨会论文集,北京:人民交通出版社,1997.
[8]朱小燕.手写体字符识别研究[J].模式识别与人工智能,2000(2):173-179.
[9] Kenneth R Castleman.数字图像处理[M],北京:北京电子工业出版社,2004.
[10]李小平.车辆牌照识别系统可靠性问题的研究[J].北京理工大学学报,2001,21(2):1-14.
[11]高守传,姚领田等.Visual C++----数字图像处理与工程应用篇[M].北京:中国铁道出版社,2002:20 .
[12]于万波,沈连山,赵品.基于颜色特征的汽车车牌定位[J].辽宁工程技术大学学报(自然科学版),2001,20(5):713-714.
[13]吴冰,秦志远.自动确定图像二值化最佳阈值的新方法[J].测绘学院学报,2001(12):283-286.
[14]付忠良.图象阈值选取方法的构造[J].中国图象图形学报,2000(6):466-469.
[15]杨枝灵,王开.数字图像获取处理及实践应用[M].北京:人民邮电出版社,2003:535-536..
[16]叶萝芸等.文本图像的快速二值化方法[J].红外与毫米波学报,1997,16(5):344-350
[17]阮秋琦.数字图像处理基础[M].北京:中国铁道出版社,1988.
[18]徐飞,施晓红.MATLAB应用图像处理[M].西安:西安电子科技大学出版社,2002.
[19]杨海廷.基于纹理特征的车牌识别系统的研究与实现[D].成都:电子科技大学,2005.
[20]王昱,赵正校,杨硕.基于直线边缘识别的图像区域定位方法[J].计算机工程,1999,25(9):61-62.
[21]左奇,史忠科,一种基于数学形态学的实时车牌图象分割方法[J].中国图象图形学报,2003,8(3):281-285.
[22]边肇棋,张学工.模式识别[M].北京:清华大学出版社,2000.
[23]王时标,姚振兴.数域上的模糊模式识别方法[J].模糊系统与数学,1995,9(4):32-38.
[24]游素亚.图像边缘检测技术的发展与现状[J].电子科技导报,1995(8):25-28 .
[25] Rosenfeld,A C Kak.Digital picture Processing[J].New York Academic,1977 :133-170.
[26]罗文村.现代计算机[M].北京:中国铁道出版社,2001:43-46.
[27] Tremeau A , Borel N.A region growing and merging algorithm to color segmentation[J].Pattern Recognition,1997,30(7):1191-1203.
[28]赵锋等.分裂-合并方法在图像分割、目标提取中的应用[J].西北工业大学学报,2000,18(1):116-120.
[29]崔屹.图像处理与分析-数学形态学方法及应用[M].北京:北京科学出版社,2000.
[30]路小波,张光华.基于二值图像的车牌精确定位方法[J].东南大学学报(自然科学版),2005,35(6):972-974.
[31] Wen C-Y Yu C-C , Hun Z-D.A3-D Translrnrnalion to Irnhrove the Legihilily of Lieense Plale Numbers[J].Journal of Forensie Seienees,2002,470(3):578-585.
[32]富煌清,沈巍,黄心哗.汽车牌照的提取方法研究[J].模式识别与人工智能,2000,43(3):345-348.
[33]郝永杰,刘文耀,路烁.畸变汽车牌照图像的空间校正[J].西南交通大学学报,2002,37(4):417-420.
[34] Hans A Hegl.et al. AHigh Perlbnnanee Lieense Plale Reeognition System [C].IEEE International Conefrenee,Man Systems. And Cybernetiesr,1998.
[35]权伟,郑南宁,贾新春.复杂背景下的车辆牌照字符提取方法研究[J].信息与控制,2002,31(l):25-29.
[36]乐宁,翁世修.基于单个字符区域的倾斜字符校正技术的研究[J],图像识别与自动化,2002,(2):l-7.
[37]吴一全,朱兆达.图像处理中阈值选取方法30年[J].数据采集与处理,1993,8(3):193-201.
[38] YingDai,NnnaingZheng,XiningZhnag,GuorongXuan.AutomatieereognitionofProvineenmaeonthelicensePlateofmovingvehicle.gthInternationalConefreneeonPattemReeognition,1998(2):927-929.
[39]郑南宁,张西宁,戴莹,朱海安.行驶车辆牌照自动识别系统[J].西安交通大学学报,1991,l:43-53.
[40]吴传孙,邹杨德,周定康.二值化算法在车牌识别中的应用[J].计算机与现代化,2003,(6):13-18.
[41]明德烈,柳健.一种改进的手写汉字文本切分算法[J].华中理工大学学报,2000,28(2):87-89.
[42]吴大勇,魏平,侯朝桢,刘永信.一种车牌图像中的字符快速分割与识别算法[J].计算机工程与应用,2003,(3):212-215.
[43] Yanamura Y.,Goto M.,Nishiyama D,Soga M,Nakatani H,Saji H.Extraction and tracking of the license plate using Hough transform and voted block matching.IEEE IV2003 Intelligent Vehicles Symposium.2003:243-246.
[44]吴进军,杜树新.车牌字符分割新方法[J].工业控制计算机,2005,8(4):69-75.
[45]罗文村.基于阈值法与区域生长法综合集成的图像分割法[J].现代计算机2001,(5):43-47.
[46] Sang U k Lee,Seok Yook Chung.A Comparative Performance Study of Several Global Thresholding Techniques for Segmentation.Computer Vision,Graphics,and Image Understanding.1990,50(2):17-19.
[47] Zunino,Rodolfo,Rovetta,Stefano.Visual location of license plates by vector System's quantization.Proceedings IEEE International Symposium on Circuits and systems.1999:135-138.
[48]袁崇明,陈环宇.一种识别汽车牌照的字符分割技术[J].计算机时代,2004(8):5-10.
[49]任明亮,范勇,蒋欣荣,游志胜.一种快速精确的汽车牌照字符切分算法[J].四川大学学报,2002,39(3):450-454.
[50]朱小燕,手写体字符识别研究[J].模式识别与人工智能,2000,(2) :173-179.
[51]昊佑寿等.汉字识别-原理方法与实现第一版[M].北京:高等教育出版社,1992:160-181.
[52] Shi Huihan.On the Diferent Processing of Wholes and Parts.A Psychophysiological Analysis.Journal of Cognitive Neuroscience.1997,9(5):680-697.
[53]王时标,姚振兴,数域上的模糊模式识别方法[J],模糊系统与数学,1995,9(4):32-38.
[54]金连文,徐秉铮.基于多级神经网络结构的手写体汉字识别[J].通信学报,1997,18(5):21-27.
[55] Kenneth R Castleman,数字图像处理[M].北京:北京电子工业出版社,2004.
[56]刘新宇.汽车牌照识别系统应用研究[D].郑州:控制工程学院,2004.
[57]章毓晋.图象工程(上册)—图象处理和分析.第1版,北京:清华大学出版社,1999.
[58] Sunghoon Kim,Daeehul Kimetal“ARobust License-Plate Extraction Methold under Complex Image Conditions”16th ICPR IEEE,Quebec,Canada,2002P.P:216-219。
[59]张学工.关于统计学习理论与支持向量机[J].自动化学报,2000,26(l):32-42.
[60] Defries,R.S.,Townshend,J.R.G NDVI-derived L and Cover Classificationata Global Seale[J].International Joumal of Re-mote Sensing,1994,15(17):3567-3586.