车牌的定位与识别
|
摘要
车辆牌照自动识别系统是近几年发展起来的基于图形图像处理和字符识别术的智能化交通管理系统的一个核心组成部分,是目前世界范围内模式识别研究领域的一个热点。本文对车辆牌照定位及识别系统中的图像预处理、字符分割和字符识别等技术所涉及的新算法、新设计做了一个比较详细的论述。
本文在图像预处理中采用了灰度图像二值化的算法和基于数学形态学的图像去除噪声的方法。基于数学形态学的图像去除噪声是通过对图像的开、闭操作有选择的去噪。可以去除直径小于字符笔划半径的孤立噪声点。
本文在研究了车牌定位及分割的一些经典方法的基础上,采用基投影法的定位方法实现车牌的定位,利用了灰度投影法和连通域分析综合方法最终较好地实现了车牌字符的分割。
本文重点对基于BP神经网络的车辆牌照识别技术进行了深入的研究和分析。首先介绍了汽车牌照识别系统的现状;其次对神经网络的构成以及相关的理论进行了讨论,着重分析了神经网络的理论原理和在字符识别中的应用方法并给出了实验结果。
本论文结论如下:
1、研究了车牌区域定位模块所涉及到一些预处理方法。针对车牌图象对比度增强的特点,设计了一段灰度变换函数,探讨了车牌图像定位模块所涉及到的车牌图像二值化技术。
2、对车牌定位分割的一些经典方法进行了分类归纳,设计并实现了基于投影法的定位分割方法,设计的方法具有分割比较准确、运算简单的优点,适合实际使用。
3、本论文利用神经网络技术,对车牌自动识别系统进行研究。在所拍摄的车牌照片中,先获取车牌位置,再把车牌的各个字符分割,提取出各个字符的特征,再利用BP神经网络进行识别,从而获取车牌号码。
The technology of the automatic vehicle identification is one of the important studies in the intelligence traffic system.License Plate Recognition is the organic part of the Intelligent Transportation System(ITS),which play a key role in the ITS.One function of the ITS is achieve the aim of automatic control and manage.Then the License Plate Recognition can provide the needed vehicle identification information for the subsystems of the ITS.With those it can accelerate the automatic manage of the traffic system.LPR can be used during the crossways,carbarn manage、big park、highway、the charge system of the bridge and so on.It is the key technology to solve the traffic bottleneck problems that are made by the traditional manage mode during the modern society.With more fast development of transportation and the use of higher-tech in the automatic monitor and control,the AVI seems more important.In this thesis,we did a lot of research on LPR key technology based on image aimed at the requirements.
The LPR system is composed with vehicle Plate capture,license Plate recognition module,database managing and data transfer through net etc..This Paper studies on the algorithms of license Plate recognition module.The module is composed with image Per-Process,Plate location,char segmentation and char recognition.
Works have been done in this paper:1.Research on LPR principle;2.Research on the Plate Location and segmentation method through Edge Detection and region search;3.Research on the Plate Location pre-process method;4.Research on character isolates method.Following are the main Work.1.The PLR system principle and universal PLR are analyzed,the following works are accomplished:Ⅰplate location and type sense.A) Magnetic induction detection;B) Transportation microwave sensor;C) Infrared detection;D) Ultrasonic detection;ⅡAnalysis on PLRA) IC Card recognition technology;B) Bar code recognition technology;C) Recognition technology based on image comprehension;D) Recognition technology based on NN;E) Digital image vehicle recognition.2.The difficulty of plate location and segmentation which is the key technology of PLR is studied.Assorted and concluded the classic method of the plate location and segmentation.Based on these the Plate Location method through Edge Detection and region search are designed and realized.During the process of the method,the paper discuss the Edge Detection、region search and plate region re-distinguish that are three key taches.ⅠAnalysis on plate location.A) Method based on linear detection;B) Method based on threshold;C) Method based on gray edge detection;D) Method based on colorful image.ⅡThe paper finished the Plate Location method through Edge Detection and region search based on upwards researches.A) Basic way of the method;B) Method realization.3. The predispose method of Plate Location module is studied.Then Subsection gray transform function is designed.Using hough transformation and image reverse rotation to rectificate slant,the following works are accomplished:ⅠPredisposal on image.A) Plate image enhancement;B) Gray transform.ⅡEquilibrium disposal.ⅢSlant rectification and zoom disposal.ⅣPlate image binarization.4.We analyzed the classic way and method of the problem of the character isolate,designed the character horizontal and vertical segmentation method,and designed the segmentation result method to boost the veracity.ⅠHorizontal segmentation method based on edge detection;ⅡVertical segmentation based on projection and knowledge;A) Calculate vertical projection;B)Apices-vale inspection method principal vertical segmentation;C) Character split and unites;D) Wipe off the forward and back disturbs.
Main achievement are accomplished in this paper:1.We present a new method, which is mufti-vehicle plates segmentation and recognition.First,we use the statistics and the feature extraction to extract the background,which contain the vehicle license plates.Second,after making sure of the location of vehicle license plates,characters in the vehicle license are divided and the location of the characters is located.At last, we use the BAM to recognize the vehicle license precisely.2.Designing and realizing method through Edge Detection and region search.During the realizing process of the method,the paper discuss the Edge Detection、region search and plate region re-distinguish that are three key taches.3.Subsection gray transform function is designed.4.Horizontal and vertical segmentation method is designed.5.Syncopate result rectification tache are designed to improve the veracity of character syncopate.
Further Research Works can be continued further as followed:
1.For people having researched the LPR for many years and the time limitation, we take the image manipulation as the method without theory innovation.
2.The LPR did not use the colorful information and detect on the gray image, next we can bring colorful information to improve the effect of plate location and character binarization.
3.Improving the toughness of character and recognition probability to deformity、empoison and distortion,next our work is to seek character which is more steady and choose more effective class sensor.
4.The recognition probability of Chinese characters is lower than letter and there are more superior room.
本文在图像预处理中采用了灰度图像二值化的算法和基于数学形态学的图像去除噪声的方法。基于数学形态学的图像去除噪声是通过对图像的开、闭操作有选择的去噪。可以去除直径小于字符笔划半径的孤立噪声点。
本文在研究了车牌定位及分割的一些经典方法的基础上,采用基投影法的定位方法实现车牌的定位,利用了灰度投影法和连通域分析综合方法最终较好地实现了车牌字符的分割。
本文重点对基于BP神经网络的车辆牌照识别技术进行了深入的研究和分析。首先介绍了汽车牌照识别系统的现状;其次对神经网络的构成以及相关的理论进行了讨论,着重分析了神经网络的理论原理和在字符识别中的应用方法并给出了实验结果。
本论文结论如下:
1、研究了车牌区域定位模块所涉及到一些预处理方法。针对车牌图象对比度增强的特点,设计了一段灰度变换函数,探讨了车牌图像定位模块所涉及到的车牌图像二值化技术。
2、对车牌定位分割的一些经典方法进行了分类归纳,设计并实现了基于投影法的定位分割方法,设计的方法具有分割比较准确、运算简单的优点,适合实际使用。
3、本论文利用神经网络技术,对车牌自动识别系统进行研究。在所拍摄的车牌照片中,先获取车牌位置,再把车牌的各个字符分割,提取出各个字符的特征,再利用BP神经网络进行识别,从而获取车牌号码。
The technology of the automatic vehicle identification is one of the important studies in the intelligence traffic system.License Plate Recognition is the organic part of the Intelligent Transportation System(ITS),which play a key role in the ITS.One function of the ITS is achieve the aim of automatic control and manage.Then the License Plate Recognition can provide the needed vehicle identification information for the subsystems of the ITS.With those it can accelerate the automatic manage of the traffic system.LPR can be used during the crossways,carbarn manage、big park、highway、the charge system of the bridge and so on.It is the key technology to solve the traffic bottleneck problems that are made by the traditional manage mode during the modern society.With more fast development of transportation and the use of higher-tech in the automatic monitor and control,the AVI seems more important.In this thesis,we did a lot of research on LPR key technology based on image aimed at the requirements.
The LPR system is composed with vehicle Plate capture,license Plate recognition module,database managing and data transfer through net etc..This Paper studies on the algorithms of license Plate recognition module.The module is composed with image Per-Process,Plate location,char segmentation and char recognition.
Works have been done in this paper:1.Research on LPR principle;2.Research on the Plate Location and segmentation method through Edge Detection and region search;3.Research on the Plate Location pre-process method;4.Research on character isolates method.Following are the main Work.1.The PLR system principle and universal PLR are analyzed,the following works are accomplished:Ⅰplate location and type sense.A) Magnetic induction detection;B) Transportation microwave sensor;C) Infrared detection;D) Ultrasonic detection;ⅡAnalysis on PLRA) IC Card recognition technology;B) Bar code recognition technology;C) Recognition technology based on image comprehension;D) Recognition technology based on NN;E) Digital image vehicle recognition.2.The difficulty of plate location and segmentation which is the key technology of PLR is studied.Assorted and concluded the classic method of the plate location and segmentation.Based on these the Plate Location method through Edge Detection and region search are designed and realized.During the process of the method,the paper discuss the Edge Detection、region search and plate region re-distinguish that are three key taches.ⅠAnalysis on plate location.A) Method based on linear detection;B) Method based on threshold;C) Method based on gray edge detection;D) Method based on colorful image.ⅡThe paper finished the Plate Location method through Edge Detection and region search based on upwards researches.A) Basic way of the method;B) Method realization.3. The predispose method of Plate Location module is studied.Then Subsection gray transform function is designed.Using hough transformation and image reverse rotation to rectificate slant,the following works are accomplished:ⅠPredisposal on image.A) Plate image enhancement;B) Gray transform.ⅡEquilibrium disposal.ⅢSlant rectification and zoom disposal.ⅣPlate image binarization.4.We analyzed the classic way and method of the problem of the character isolate,designed the character horizontal and vertical segmentation method,and designed the segmentation result method to boost the veracity.ⅠHorizontal segmentation method based on edge detection;ⅡVertical segmentation based on projection and knowledge;A) Calculate vertical projection;B)Apices-vale inspection method principal vertical segmentation;C) Character split and unites;D) Wipe off the forward and back disturbs.
Main achievement are accomplished in this paper:1.We present a new method, which is mufti-vehicle plates segmentation and recognition.First,we use the statistics and the feature extraction to extract the background,which contain the vehicle license plates.Second,after making sure of the location of vehicle license plates,characters in the vehicle license are divided and the location of the characters is located.At last, we use the BAM to recognize the vehicle license precisely.2.Designing and realizing method through Edge Detection and region search.During the realizing process of the method,the paper discuss the Edge Detection、region search and plate region re-distinguish that are three key taches.3.Subsection gray transform function is designed.4.Horizontal and vertical segmentation method is designed.5.Syncopate result rectification tache are designed to improve the veracity of character syncopate.
Further Research Works can be continued further as followed:
1.For people having researched the LPR for many years and the time limitation, we take the image manipulation as the method without theory innovation.
2.The LPR did not use the colorful information and detect on the gray image, next we can bring colorful information to improve the effect of plate location and character binarization.
3.Improving the toughness of character and recognition probability to deformity、empoison and distortion,next our work is to seek character which is more steady and choose more effective class sensor.
4.The recognition probability of Chinese characters is lower than letter and there are more superior room.
引文
[1]牛欣,沈兰荪.汽车牌照识别技术的研究与测控技术[J],1999,18(12):14-赵永进.我国智能交通系统现状与发展趋势公安交通科技窗.2002年第5期:7-10.
[2]张立明.人工神经网络的模型及其应用.上海:复旦大学出版1993:24-56.
[3]RohlfingT,RussakoffDB,MaurerCR.Performanee-based classifier combination in Atlas-based image segmentation using expectation-maximization Parameterestimation[J].IEEETransaetionsonMediealImaging,2004,23:83-94.
[4]张宏林。VisualC++数字图像模式识别技术及工程实践[M].人民邮电出版社,2003.2:220-221.
[5]HongliangB,ChanpingL.Ahybrid license plate extraction method based on edge statisties and morphology[J],patten Recognition,2004 Proceedings of the 17th Intemational Conference on ICPR 2004.VolumeZ,23-26Ang.2004:831-834.
[6]邢博,梁得群,李文举,一种新的车牌数字及字母字符识别方法[J].辽宁师范大学学报(自然科学版)2005.3:56-58.
[7]王春,刘波,周新志,采用BP神经网络的车牌字符识别方法研究[J].中国测试技术,2005.1:26-29.
[8]张引、潘云鹤.彩色图象定位新方法[J],中国图象图形学报,2001,6A(4):374-377.
[9]魏波,图像日标的识别模型与检测技术.光电工程,1999(6).15-18.
[10]郭捷,施鹏飞.基于颜色和纹理分析的车牌定位方法中国图象图形学报,2002,7A(5):472-475.
[11]KAMAT V,GANESA S.An efficient implementation of the Plough transform for detecting vechicle license plates using DSPpS Proceeding of Real-Time Technology and Applications Symposium.CA IEEE Computer Society Press,1995:58-59.
[12]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中国图象图形学报,2003,8A (5):281-295.
[13]王栓,艾海舟,何克忠.基于差分图象的多运动目标的检测与跟踪中国图象图形学报,1999 6 vol4(A) No 6.
[14]曹辿铭,宜国荣.牌照字符分割中的区域分裂与合并.计算机工程,2000,26(10):23-27.
[15]许健,段小梅.车牌定位技术在“电子警察”系统中的应用研究.自动化与信息工程,2006,27(3):18-20.
[16]王润生.图像理解.北京:国防科技大学出版社,1995:156-159.
[17]D.S.Kim,S.I.Chien.Automatic Car License Plate Extraction Using Modified Generalized Symmetry Transform and Image Warping.IEEE International Symposium on Industrial Electronics,2001,3:2022-2027.
[18]Christos Nikolaos E.AnagnostoPoulos,Vassili Loumos.A License Plate-Recognition Algorithm for Intelligent Transportation System Applications.IEEE Transactions on Intelligent Transportation Systems,2006,7(3):377-391.
[19]何斌,马天宇,王运坚,朱红莲。VisualC++数字图像处理。人民邮电出版社,(2002):134-150.
[20]S.Nomura,K.Yamanaka,O.Katai,H.Kawakami,and T.Shiose,"A novel adaptive morphological approach for degraded charaeter image segmentation," Pattern Reeognit.,vol.38,no.11,2005:1961-1975.
[21]Y.Cui and Q.Huang,"Extraeting charaeters of license plates from video sequences," mach,Vis.Appl.,vol.10,no.5/6,1998:308-320.
[22]张旭,王宏安,戴国忠,面向车牌识别的区域分割技术[J],计算机工程,2002,28(4):113-115.
[23]陈黎,黄心汉,王敏,李伟,基于聚类分析的车牌字符分割方法[J],计算机工程与应用,2002,6:221-222,256.
[24]SalvatoreA.D' Agostino Application of Machine Vision to Electronic Toll and Traffic Management A Survey.SPIE Vol.2592,10-16.
[25]T.Takaba,M.Sakarchi,T,Kaneko Measurement of Traffic Flow Using RealTime Proeessing of Moving Pieture Proceedings 32nd IEEE conference on vehicle Teehnology.
[26]SalvatoreA.D' Agostino,Paul Shuldiner.Application of Video/machine Vision Technology in Traffic Data Analysis:ASurvey.SPIE Vol.2344 Intelligent Vehiele Highway Systems(1994),207-214.
[27]R Parisi,Car Plate Recognition by Neural Networks and Image Processing.[C].In Proe.IEEE ISCAS.USA,1998.
[28]M.H.ter Brugge etc.License Plate Recognition Using DTCNNs,1998Fifth,IEEE international workshop Cellular Neural Networks and their hPPlieations.London,England:Apri11998:14,17.
[29]赵雪春,戚飞虎。基于彩色分割的车牌自动识别技术[J]。上海交通大学学报,1998,32(10):4-9.
[30]沈会良,李志能。基于CCD的汽车牌照自动识别系统[J]。光电工程,2000,27(4):60-63,71.
[31]张玉芝,史忠科。一种新的车牌识别与处理算法[J]。西北工业大学学报,2002,20(1):83、86.
[32]T.D.Duan,T.L.HongDu,T.V.Phuoc,and N.V.Hoang," Building an automatic vehicle license plate recognition system,"in Proe.Int.Conf.ComPut.Sei.RIVF,2005:59-63.
[33) C.Anagnostopoulos,E.Kayafas,andV.Loumos.(2000).Digital image Processing and neural networks for vehicle license Plate identifieation.J.Eleet.Eng.[online].1(2):2-7.Available:http://~.medialab.ntua.gr/veople/eanag/journ als.php.
[34]F.Martin,M.Gareia,and L.Alba," New methods for automatic reading of VLP's(VehieleLieensePlates)," in Proc IASTED Int.Conf.SPPRA,Jun.2002.[Online].Available:http.//www.gpi.tse.uvigo.es/pub/papers/sPPra02.Pdf.
[35]K.K.Kim,K.1.Kim,J.B.Kim,and H.J.Kim,"Learning- based approch,for license plate recognition," inProc.IEEESignalProeess.Soe.Workshop,NeuralNetw.Signal Proeess.,2000,vol.2,pp.614-623.
[36]P.Comelli,P.Ferragina,M.N.Granieri,and F.Stabile," Optical recognition of motor vehicle license plates," IEEE Trans.Veh.Zeehnol.,vol.44,no.4,PP.790-799,Nov.1995.
[2]张立明.人工神经网络的模型及其应用.上海:复旦大学出版1993:24-56.
[3]RohlfingT,RussakoffDB,MaurerCR.Performanee-based classifier combination in Atlas-based image segmentation using expectation-maximization Parameterestimation[J].IEEETransaetionsonMediealImaging,2004,23:83-94.
[4]张宏林。VisualC++数字图像模式识别技术及工程实践[M].人民邮电出版社,2003.2:220-221.
[5]HongliangB,ChanpingL.Ahybrid license plate extraction method based on edge statisties and morphology[J],patten Recognition,2004 Proceedings of the 17th Intemational Conference on ICPR 2004.VolumeZ,23-26Ang.2004:831-834.
[6]邢博,梁得群,李文举,一种新的车牌数字及字母字符识别方法[J].辽宁师范大学学报(自然科学版)2005.3:56-58.
[7]王春,刘波,周新志,采用BP神经网络的车牌字符识别方法研究[J].中国测试技术,2005.1:26-29.
[8]张引、潘云鹤.彩色图象定位新方法[J],中国图象图形学报,2001,6A(4):374-377.
[9]魏波,图像日标的识别模型与检测技术.光电工程,1999(6).15-18.
[10]郭捷,施鹏飞.基于颜色和纹理分析的车牌定位方法中国图象图形学报,2002,7A(5):472-475.
[11]KAMAT V,GANESA S.An efficient implementation of the Plough transform for detecting vechicle license plates using DSPpS Proceeding of Real-Time Technology and Applications Symposium.CA IEEE Computer Society Press,1995:58-59.
[12]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中国图象图形学报,2003,8A (5):281-295.
[13]王栓,艾海舟,何克忠.基于差分图象的多运动目标的检测与跟踪中国图象图形学报,1999 6 vol4(A) No 6.
[14]曹辿铭,宜国荣.牌照字符分割中的区域分裂与合并.计算机工程,2000,26(10):23-27.
[15]许健,段小梅.车牌定位技术在“电子警察”系统中的应用研究.自动化与信息工程,2006,27(3):18-20.
[16]王润生.图像理解.北京:国防科技大学出版社,1995:156-159.
[17]D.S.Kim,S.I.Chien.Automatic Car License Plate Extraction Using Modified Generalized Symmetry Transform and Image Warping.IEEE International Symposium on Industrial Electronics,2001,3:2022-2027.
[18]Christos Nikolaos E.AnagnostoPoulos,Vassili Loumos.A License Plate-Recognition Algorithm for Intelligent Transportation System Applications.IEEE Transactions on Intelligent Transportation Systems,2006,7(3):377-391.
[19]何斌,马天宇,王运坚,朱红莲。VisualC++数字图像处理。人民邮电出版社,(2002):134-150.
[20]S.Nomura,K.Yamanaka,O.Katai,H.Kawakami,and T.Shiose,"A novel adaptive morphological approach for degraded charaeter image segmentation," Pattern Reeognit.,vol.38,no.11,2005:1961-1975.
[21]Y.Cui and Q.Huang,"Extraeting charaeters of license plates from video sequences," mach,Vis.Appl.,vol.10,no.5/6,1998:308-320.
[22]张旭,王宏安,戴国忠,面向车牌识别的区域分割技术[J],计算机工程,2002,28(4):113-115.
[23]陈黎,黄心汉,王敏,李伟,基于聚类分析的车牌字符分割方法[J],计算机工程与应用,2002,6:221-222,256.
[24]SalvatoreA.D' Agostino Application of Machine Vision to Electronic Toll and Traffic Management A Survey.SPIE Vol.2592,10-16.
[25]T.Takaba,M.Sakarchi,T,Kaneko Measurement of Traffic Flow Using RealTime Proeessing of Moving Pieture Proceedings 32nd IEEE conference on vehicle Teehnology.
[26]SalvatoreA.D' Agostino,Paul Shuldiner.Application of Video/machine Vision Technology in Traffic Data Analysis:ASurvey.SPIE Vol.2344 Intelligent Vehiele Highway Systems(1994),207-214.
[27]R Parisi,Car Plate Recognition by Neural Networks and Image Processing.[C].In Proe.IEEE ISCAS.USA,1998.
[28]M.H.ter Brugge etc.License Plate Recognition Using DTCNNs,1998Fifth,IEEE international workshop Cellular Neural Networks and their hPPlieations.London,England:Apri11998:14,17.
[29]赵雪春,戚飞虎。基于彩色分割的车牌自动识别技术[J]。上海交通大学学报,1998,32(10):4-9.
[30]沈会良,李志能。基于CCD的汽车牌照自动识别系统[J]。光电工程,2000,27(4):60-63,71.
[31]张玉芝,史忠科。一种新的车牌识别与处理算法[J]。西北工业大学学报,2002,20(1):83、86.
[32]T.D.Duan,T.L.HongDu,T.V.Phuoc,and N.V.Hoang," Building an automatic vehicle license plate recognition system,"in Proe.Int.Conf.ComPut.Sei.RIVF,2005:59-63.
[33) C.Anagnostopoulos,E.Kayafas,andV.Loumos.(2000).Digital image Processing and neural networks for vehicle license Plate identifieation.J.Eleet.Eng.[online].1(2):2-7.Available:http://~.medialab.ntua.gr/veople/eanag/journ als.php.
[34]F.Martin,M.Gareia,and L.Alba," New methods for automatic reading of VLP's(VehieleLieensePlates)," in Proc IASTED Int.Conf.SPPRA,Jun.2002.[Online].Available:http.//www.gpi.tse.uvigo.es/pub/papers/sPPra02.Pdf.
[35]K.K.Kim,K.1.Kim,J.B.Kim,and H.J.Kim,"Learning- based approch,for license plate recognition," inProc.IEEESignalProeess.Soe.Workshop,NeuralNetw.Signal Proeess.,2000,vol.2,pp.614-623.
[36]P.Comelli,P.Ferragina,M.N.Granieri,and F.Stabile," Optical recognition of motor vehicle license plates," IEEE Trans.Veh.Zeehnol.,vol.44,no.4,PP.790-799,Nov.1995.