欢迎来到天天文库
浏览记录
ID:36804240
大小:5.07 MB
页数:61页
时间:2019-05-15
《一种基于RBF神经网络的车牌识别技术的研究》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、一种基于RBF神经网络的车牌识别技术的研究摘要车牌识别是实现交通智能化的核心技术之一,在智能交通系统领域有着重要的应用价值,车牌识别技术的运用,将大大改善公路交通运行和管理的效率。车牌自动识别系统一般包括车牌定位、字符分割和字符识别三个部分。车牌定位是车牌识别系统处理的第一步,定位的准确与否直接关系着车牌识别的成败。本文中先对图像进行预处理,然后进行数学形态学操作,得到若干个候选区域,最后根据车牌的特征选取真正的车牌区域。车牌字符的切分效果直接影响到字符识别,在字符切分之前先进行倾斜校正及二值化的处理,然后采用基于先验知识
2、的行扫描法对车牌进行去除边框、铆钉的预处理。在切分方面本文采用了一种利用垂直投影信息结合先验知识约束的垂直分割方法,并对算法进行了改进,有效地提高了字符切分的准确性。车牌的字符识别是整个车牌识别系统的核心部分,本文采用RBF神经网络模式识别技术作为识别的方法,构造了3个子网络进行车牌字符的识别,并采用一种改进的训练方法,有效的提高了RBF神经网络的识别性能。实验结果表明,本系统能较准确定位、分割车牌并进行识别,系统的性能良好。从中可看出:多种预处理与识别技术有机结合能提高系统识别能力。关键词:车牌识别图像预处理数学形态学径
3、向基神经网络ASTUDYOFTECHNOLOGYINLICENSEPLATERECOGNITIONBASEDONTHERADIALBASISFUNCTl0NNEI瓜AI。NETWO砒4、vethetrafficmoveandmanagementemciency.LPRsystemconsistsofthreemodulesingeneral,thoseare:1icenseplatelocating,charactersegmentationandcharacterrecognition.LocatethePlateiSthefaststepoftheLPRsystem;itsaccuracydirectlyaffectstheSUCCESSofmecharacterrecognitionresult.F5、irstly,imagepreprocessingisadopted,whichisthenfollowedbymathematicalmorphologyoperations,secondly,somepreparativeareasareobtained,andf'mally,thereallicenseplateareaisobtainedaccordingtothecharacteristicofthelicenseplate.Theresultofcharactersegmentationaffectsthere6、cognitiondirectly,firstly,slant-correctionandbinarizationareprocessed,then,aline—scanningmethodbasedontranscendentknowledgeofplatetoistakentowipeofftheframesandrivets.Inthepaper,themethodofcharacterssegmentationiSconnectedverticalprojectioninformationandthetransce7、ndentknowledge,andthenimprovesthealgorithm,whichimprovestheaccuracyofsegmentationeffectively.CharacterrecognitioniStheverycoreoftheLPRsystem.InthethesiswestructurethreenetworkstoapplytocharactersofLPRbasedonRadialBasisFunctionNeuralNetworks(P,BFrCN).Throughthefurt8、hertrainingmethod,therecognitioncapabilityoftheRB删iSimprovedmuchmore.Asthetestresult,itprovesthatthisrecognitionsystemcanrelativelyaccuratetolocatelicen
4、vethetrafficmoveandmanagementemciency.LPRsystemconsistsofthreemodulesingeneral,thoseare:1icenseplatelocating,charactersegmentationandcharacterrecognition.LocatethePlateiSthefaststepoftheLPRsystem;itsaccuracydirectlyaffectstheSUCCESSofmecharacterrecognitionresult.F
5、irstly,imagepreprocessingisadopted,whichisthenfollowedbymathematicalmorphologyoperations,secondly,somepreparativeareasareobtained,andf'mally,thereallicenseplateareaisobtainedaccordingtothecharacteristicofthelicenseplate.Theresultofcharactersegmentationaffectsthere
6、cognitiondirectly,firstly,slant-correctionandbinarizationareprocessed,then,aline—scanningmethodbasedontranscendentknowledgeofplatetoistakentowipeofftheframesandrivets.Inthepaper,themethodofcharacterssegmentationiSconnectedverticalprojectioninformationandthetransce
7、ndentknowledge,andthenimprovesthealgorithm,whichimprovestheaccuracyofsegmentationeffectively.CharacterrecognitioniStheverycoreoftheLPRsystem.InthethesiswestructurethreenetworkstoapplytocharactersofLPRbasedonRadialBasisFunctionNeuralNetworks(P,BFrCN).Throughthefurt
8、hertrainingmethod,therecognitioncapabilityoftheRB删iSimprovedmuchmore.Asthetestresult,itprovesthatthisrecognitionsystemcanrelativelyaccuratetolocatelicen
此文档下载收益归作者所有