多類型車牌識別配置的方法畢業(yè)論文外文翻譯

1、一、英文原文：一、英文原文：AconfigurablemethodfmultistylelicenseplaterecognitionAutomaticlicenseplaterecognition(LPR)hasbeenapracticaltechniqueinthepastdecades.Numerousapplicationssuchasautomatictollcollectioncriminalpursuittrafficla

2、wenfcementhavebeenbenefitedfromit.AlthoughsomenoveltechniquesfexampleRFID(radiofrequencyidentification)WSN(wirelesssenswk)etc.havebeenproposedfcarIDidentificationLPRonimagedataisstillanindispensabletechniqueincurrentinte

3、lligenttransptationsystemsfitsconveniencelowcost.LPRisgenerallydividedintothreesteps:licenseplatedetectionactersegmentationacterrecognition.ThedetectionsteproughlyclassifiesLPnonLPregionsthesegmentationstepseparatesthesy

4、mbolsactersfromeachotherinoneLPsothatonlyaccurateoutlineofeachimageblockofactersisleftftherecognitiontherecognitionstepfinallyconvertsgreylevelimageblockintoacterssymbolsbypredefinedrecognitionmodels.AlthoughLPRtechnique

5、hasalongresearchhistyitisstilldrivenfwardbyvariousarisingdemsthemostfrequentoneofwhichisthevariationofLPstylesfexample:(1)Appearancevariationcausedbythechangeofimagecapturingconditions.(2)Stylevariationfromonenationtoano

6、ther.(3)StylevariationwhenthegovernmentreleasesnewLPfmat.WesummedthemupintofourfactsnamelyrotationanglelinenumberactertypefmataftercomprehensiveanalysesofmultistyleLPacteristicsonrealdata.Generallyspeakinganychangeofthea

7、bovefourfactscanresultinthechangeofLPstyleappearancethenaffectthedetectionsegmentationrecognitionalgithms.IfoneLPhasalargerotationanglethesegmentationrecognitionalgithmsfhizontalLPmaynotwk.Iftherearemethanoneacterlinesin

8、oneLPadditionallineseparationalgithmisneededbefeasegmentationprocess.Withthevariationofactertypeswhenweapplythemethodfromonenationtoanothertheabilitytoredefinetherecognitiondetection.InRef.Kimetal.usedanSVMtotraintexture

9、classifierstodetectimageblockthatcontainsLPpixels.InRef.theauthsusedGabfilterstoextracttexturefeaturesinmultiscalesmultiientationstodescribethetexturepropertiesofLPregions.InRef.ZhangusedXYderivativefeaturesgreyvaluevari

10、anceAdaboostclassifiertoclassifyLPnonLPregionsinanimage.InRefs.waveletfeatureanalysisisappliedtoidentifyLPregions.Despitethegoodperfmanceofthesemethodsthecomputationcomplexitywilllimittheirusability.Inadditiontexturebase

11、dalgithmsmaybeaffectedbymultilingualfacts.MultilineLPsegmentationalgithmscanalsobeclassifiedintothreeclassesnamelyalgithmsbasedonprojection，binarizationglobaloptimization.Intheprojectionalgithmsgradientcolprojectiononver

12、ticalientationwillbecalculatedatfirst.The“valleys”ontheprojectionresultareregardedasthespacebetweenactersusedtosegmentactersfromeachother.Segmentedregionsarefurtherprocessedbyverticalprojectiontoobtainpreciseboundingboxe

13、softheLPacters.SincesimplesegmentationmethodsareeasilyaffectedbytherotationofLPsegmentingtheskewedLPbecomesakeyissuetobesolved.Inthebinarizationalgithmsgloballocalmethodsareoftenusedtoobtainfegroundfrombackgroundthenregi

14、onconnectionoperationisusedtoobtainacterregions.Inthemostrecentwklocalthresholddeterminationslidewindowtechniquearedevelopedtoimprovethesegmentationperfmance.Intheglobaloptimizationalgithmsthegoalisnottoobtaingoodsegment

15、ationresultfindependentactersbuttoobtainacompromiseofacterspatialarrangementsingleacterrecognitionresult.HiddenMarkovchainhasbeenusedtofmulatethedynamicsegmentationofactersinLP.Theadvantageofthealgithmisthattheglobalopti

16、mizationwillimprovetherobustnesstonoise.thedisadvantageisthatprecisefmatdefinitionisnecessarybefeasegmentationprocess.actersymbolrecognitionalgithmsinLPRcanbecategizedintolearningbasedonestemplatematchingones.Fthefmerone

17、artificialneuralwk(ANN)isthemostlyusedmethodsinceitisprovedtobeabletoobtainverygoodrecognitionresultgivenalargetrainingset.AnimptantfactintraininganANNrecognitionmodelfLPistobuildreasonablewkstructurewithgoodfeatures.SVM