资源描述：
《Design method of ARM based embedded iris recognition system》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。
DesignmethodofARMbasedembeddedirisrecognitionsystemaabaWANGYuanboHEYuqingHOUYushiLIUTingaDepartmentofOpto-ElectronicEngineering,BeijingInstituteofTechnology,Beijing,P.R.China,100081bSmartirisBiometricsCo.Ltd.,Beijing,P.R.China,100081ABSTRACTWiththeadvantagesofnon-invasiveness,uniqueness,stabilityandlowfalserecognitionrate,irisrecognitionhasbeensuccessfullyappliedinmanyfields.Uptonow,mostoftheirisrecognitionsystemsarebasedonPC.However,aPCisnotportableanditneedsmorepower.Inthispaper,weproposedanembeddedirisrecognitionsystembasedonARM.Consideringtherequirementsofirisimageacquisitionandrecognitionalgorithm,weanalyzedthedesignmethodoftheirisimageacquisitionmodule,designedtheARMprocessingmoduleanditsperipherals,studiedtheLinuxplatformandtherecognitionalgorithmbasedonthisplatform,finallyactualizedthedesignmethodofARM-basedirisimagingandrecognitionsystem.ExperimentalresultsshowthattheARMplatformweusedisfastenoughtoruntheirisrecognitionalgorithm,andthedatastreamcanflowsmoothlybetweenthecameraandtheARMchipbasedontheembeddedLinuxsystem.It’saneffectivemethodofusingARMtoactualizeportableembeddedirisrecognitionsystem.Keywords:Embeddedsystem,ARM,Linux,Imageacquisition,Irisrecognition1INTRODUCTIONTraditionalwaysforpersonalidentificationdependonexternalthingssuchaskeys,passwords,etc.Butsuchthingsmaybelostorforgotten.Onepossiblewaytosolvetheseproblemsisthroughbiometrics,foreverypersonhashisspecial[1]biometricfeaturesdefinitely.Biometricsidentificationhasgainedincreasingattentionfromthewholeworld.Biometricsfeaturesthatcanbeusedforidentificationincludefingerprints,palmprints,handwriting,veinpattern,facial[2]characteristics,iris,andsomeothermethodssuchasvoicepattern,etc.Comparedwithotherbiometricmethods,theirisrecognitionhasthefollowingadvantages:1.Theirisimageacquisitionrequiresnophysicalcontact,soanirisidentificationsystemisnon-invasiveness.2.Sincetheirisiscreatedinanearlyrandommorphogeneticprocessduringthegestation,ithaslittleprobabilitytofindtwopeopleintheworldwhoseiristexturesareidentical.Soirisrecognitionisthemostaccuratemethodandhasthelowestfalserecognitionrate.3.Theirisrecognitionhasmorestabilitythanotherbiometricidentificationmethodsbecausetheirishasmuchmorefeaturesthanotherbiometricsanditwon’tchangeinpeople’slife.Withtheadvantagesofnon-invasiveness,uniqueness,stabilityandlowfalserecognitionrate,irisrecognitionhasbeenresearchedwidelyandhasabroadusage,suchassecurity,attendance,etc.Uptonow,mostoftheirisrecognitionsystemsarebasedonPC.However,theportabilityofPCislimitedbyitsweight,InternationalSymposiumonPhotoelectronicDetectionandImaging2007:RelatedTechnologiesandApplications,editedbyLiweiZhou,Proc.ofSPIEVol.6625,66251G,(2008)·0277-786X/08/$18·doi:10.1117/12.791217Proc.ofSPIEVol.662566251G-1 sizeandthehighpowerconsumption.Thusresultsinthattheusingofirisrecognitionisconfinedinfewfields,anditisinconvenienttouse.ThewaytogetridofthelimitofPCisusingembeddedsystem.InthispaperweproposedthedesignmethodofembeddedirisrecognitionsystembasedonARM.Weanalyzedthedesignmethodoftheirisimageacquisitionmodule,designedtheARMprocessingmoduleanditsperipherals,studiedtheLinuxplatformandtherecognitionalgorithmbasedonthisplatform.ThemainchipusedinthisdesignisaSAMSUNGS3C2410whoseCPUcoreisa16/32bitARM920TRISCprocessor.ItrunstheembeddedLinuxoperatingsystemandirisrecognitionprograms.TheSAMSUNGS3C2410hasenoughcapabilityforI/Ocontrollingandexternaldevicesmanagement.Ourexperimentalresultsshowthatthisplatformisfastenoughtodotheimagepre-processingandthefeatureextractionwhicharethemosttime-coststepsintheirisrecognitionprocedure.Itisalsoprovedthatthedatastreamcanflowsmoothlybetweendifferentpartsofthissystembyourexperiment.2SYSTEMHARDWAREDESIGNThewholesystemiscomposedbyfollowingparts:irisimageacquisitioncamerawiththeassistantmoduleincludinginfrared(IR)illuminationanddistancemeasurementmodulecontrolledbyaMCU;anARMcoreboardtoruntheLinuxsystemandirisrecognitionprogramsonit;aLCDtopreviewthecapturedimagesandgivetheusersomeindication.Thesystem’sdiagramisshowninFigure1.Inthefollowing,weillustratethedesignmethodofeachpart.ARMCoreBoardDistanceMeasurementPOWERFLASHJTAGIRIlluminationMCUS3C2410LCDCameraSDRAMEtherNETGPIOFigure1:Sketchoftheembeddedirisrecognitionsystem2.1ARMcoreboardThisboardisthecentralmoduleofthewholeembeddedrecognitionsystem.Itsmainpartsinclude:mainprocessingchip,memory,powersupplyandabundantglobalinterfaces.2.1.1MainprocessingchipThemainsignalprocessingchipusedinoursystemisaSAMSUNGS3C2410whoseCPUcoreisa16/32bitARM920TRISCprocessordesignedbyAdvancedRISCMachines,Ltd.Ithasveryrichperipheral.ThebasicdiagramofthischipanditsusefulportsinoursystemdesignareshowninFigure2.Thismainprocessingchipconnectsdevices:AnUSBcamerathroughUSBdeviceinterface;TheMCUforilluminationanddistancemeasurementcontrollingthroughUART0;Proc.ofSPIEVol.662566251G-2 Memories(NANDFlash,SDRAM,etc.)throughspecifiedcontrollinginterface;ALCDthroughLCDcontrollinginterface.ThisCPUisadoptedtorunembeddedLinuxoperatingsystemandirisrecognitionprogramsthatrunonLinux.Italsocontrolsdatatransmissionbetweendifferentpartsofthesystemthroughitsinterfaces.TheirisimagedataflowintoCPUthroughUSBport,distancevalueflowintoCPUthroughUART0port,theirisrecognitionresultisshownoutthroughLCDcontrolport.S3C2410ARM920TDistanceMeasurementAndUART0IRIlluninationNANDFlashBootLoader/RecognitionResult/StandardIrisImagesDataBusMemoryCONT.SDRAMIrisImageDataCacheUSBCameraUSBDeviceLCDDMA4ChannelDMAControllerIrisImageDataTransmissionLCDCONT.IrisRecognitionResultOutFigure2:Basicdiagramofthemainprocessingchip2.1.2MemoryTwokindsofmemoriesareadoptedinourdesign,SDRAMandFlash.TheSDRAMhasafastspeedforreadingandwriting.Itreaches1064Mbytepersecond.ButthedatastoredinSDRAMwillbelostwithoutpowersupply.TheFlashmemorycanstoredatanomatterwithorwithoutpowersupply,butitsspeedisnotasfastasSDRAM.SoweuseaSDRAMasthecache.TheNANDFlashisadoptedtostoreprogramsandtheresultofrecognition.SomeirisimagesandfeaturesarealsostoredinNANDFlash.2.1.3InterfacesPlentyofinterfacesarecontainedontheARMcoreboard,includingUSB2.0,SPI,IIC,UART,Ethernetinterface,etc.Thatmakesiteasyandconvenienttoconnectandcommunicatewithotherdevices.ThemainportsusedinoursystemisUSBandUART,whichcantransferthedataandcontrolsignalsbetweentheimageacquisitionmoduleandthemaincoreboard.2.1.4PowersupplyPowersupplyisdefinitivetowhetherthewholerecognitionsystemcanworksteadilyornot.ThissystemispoweredbyProc.ofSPIEVol.662566251G-3 a5Vvoltagesource.Butnoteverypartsofthesystemcanworkunderthisvoltage,sowemustdesignrespectivepowersupplyforeachofthem.Forembeddedirisrecognitionsystem,itmustbemadesurethatthekernelandtheI/Oportare[3]poweredatthesametime.Otherwise,thechipsmaybedamaged.Inthisdesign,MIC5207isadoptedforpowersupply.2.2TheimageacquisitionmoduleProvidinghighqualityimagesareessentialtohighaccuracyrateofrecognitionalgorithm.Wecombinetheillumination,USBcameraanddistancemeasurementasthewholemoduleforclearimageacquisition.ThebasicstructureofourcameramoduleanditsassistantcomponentsareshowninFigure3.OhD'31KFEDUL0flt1/ICnFGUVWVIC°'BOrLq8TQOACCBTT+CI/IO2IJJJ2UOLxFigure3:USBcameraandassistantunitsforimageacquisition2.2.1Illumination[4]WithInfraredillumination,theirisimagecameracangettheiristexturesmoreclearly.IRhasanotheradvantagethanvisiblelightilluminationinirisimageacquisition:itcannotbefeltbypeople’seye.Thatcanavoidcausingusersuncomfortablefeeling.Inthisdesign,weuse3groupofIRLEDsof700-900nmforillumination.ItslightandclosestatesarecontrolledbytheMCU.Throughsendingdifferentsignal,wecaneasilycontroltheLEDsaccordingtooursystem’sneed.2.2.2USBcamera[4]Inthisdesign,weusedafixedfocuslens.Usersneedmoveforwardandbackwardtoimagetheirisinfocus.Andsomeadditionalmethodsshouldbeadoptedtohelppeoplefindthebestpositiontogetanirisimagegoodenough.Thatmaybesomehownotconvenient,butsuchanopticallenswillbringgreatadvantagessuchassmallsizeandlowcost.Sincetheuseoffixedfocus,thedepthoffieldisfixed.Itisrelatedwiththelens’relativeapertureF,thediameterofpupilD,thefocallengthf,thedistancebetweentheuserandtheimagesensor,andthediameterofpermitted“blurspot”whichisdecidedbythepixelsizeoftheimagesensor.Accordingtotheserelationships,wedesignseveraldifferentlenstoimagetheuser’sirisatdifferentdistance.TheimagesensorusedinoursystemisOV7620.Itsmaximumresolutionis640×480.ThepowerconsumptionislessProc.ofSPIEVol.662566251G-4 than120mWatworkstateandlessthan12uWatstandbystate.ThemostimportantfactorofthissensoristhatithashighsensitivitytoIR.Figure4isitsmonochromeresponsecurve.Fromitwecanseeitstillhasresponsetovisiblelight.SoweaddaspecialfilterinfrontofthelensinordertogetaclearerimagingunderIRillumination.1.20MonochromeResponse1.000.800.600.400.200.0050060070080090010001100Wavelength/nmFigure4:Themonochromeresponseofimagesensor2.2.3DistancemeasurementAsthedepthfieldisfixed,theusershouldstandatacertaindistancetogettheirisimageinfocus.Forgivingtheusersomecuestomovetothespecificarearapidly,weuseanIRSHARPsensorGP2D12tomeasurethisdistance.Itcangivetheuser’sdistancepreciselyinrealtime.Ifthedistanceisnotappropriateforclearlyimaging,itcanalsosendsignalstotheprocessingalgorithmsothattheblurredimagewillbediscardedfortherecognition.AMCUisadoptedtocontrolthedistancemeasurementmodule.ThevalueofdistancebetweenuserandcamerawillbereadbythisMCU.HereweuseaMEGA8MCUtoprocessthedistancedata.ThisMCUisalsousedinilluminationcontrol.2.3DisplayATFTLCDisemployedinthisdesigntogivesomecuestotheuseranddisplaytherecognitionresult.Italsosupportsapreviewforthecameraimages.AstherehasaLCDcontrollerandtheS3C2410hasthespecialdisplayport,itisconvenienttoconnecttheLCDtotheARMcoreanddriveitdirectly.3SYSTEMPLATFORMANDSOFTWAREDESIGNWeusetheLinuxOperatingSystemtorunourrecognitionalgorithms.Wedesignandoptimizeouralgorithms,sothatithasfastspeed,strongadaptabilityandhighefficiency.Inthefollowing,basedonthesystem’sworkingflow,weillustrateourplatformandsoftwaredesign.3.1SystemworkflowTheworkflowofthissystemisshowninFigure5.Firstlyinitializethecameraandstartmeasuringthedistancebetweentheuserandthecamera.Thenreadimagedatafromcamera.Ifthedistancebetweentheuserandthecameraisappropriateforclearlyimaging,thesystemwilltranslateimagedataintoYUVthroughVideo4Linux(V4L).V4Listhekerneldriveaboutvideodevices.Itprovidesagroupofinterfacefunctionsforapplicationprogramsaboutvideo.Atlast,Proc.ofSPIEVol.662566251G-5 theirisimageisprocessedbytherecognitionalgorithmandtheresultisshowedthroughLCD.Dependingonthesystemworkflow,wedesignoursystem’ssoftware.ItiscomposedbyembeddedLinuxkernelandirisrecognitionapplicationprogram.3.2EmbeddedLinuxoperatingsystemInthisdesign,weuseaLinuxoperatingsystem(OS)asplatformforthewholerecognitionprocess.TheJUTJJJJ1oUembeddedOSappliesglobalroutineinterfaces.WiththeuseofembeddedOS,evensoftwareengineerswhoGJqjmD'T'EL0Wdonotknowmuchmoreaboutrock-bottomhardwiresD!4JUCGcanwriteprogramsforembeddedsystems.EmbeddedI/TGrrILGmGULinuxhasadvantagessuchasopensourcesoftware,JJJGflGLjrbbLobucqTwucGefficientandsteadykernel,friendlydevelopenvironment,beingcompatibletomanyhardwareXE2platforms,etc.Inthissystem,embeddedLinux1LJULOLoperatingsystemprovidesprocessmanagement,)JJmemorymanagement,documentmanagement,V4Ldataformtransformanddevicescontrol.Thesystem’ssoftwarestructureisshowninFigure6.ItcontrolsallJLKGC0U!!0Lthecommunicationbetweenthehardwaresystems,suchastheimageacquisitionanddatatransfer.Besidesthis,therecognitionalgorithmandthesoftwarealsorunonKG11J011thisplatform.Figure5:TheworkflowofthissystemJLffGCOJJJJOJJE)LOLJJJLLOCGJ/JGJJJOL;DOCIIWGUAFUGJJJUGJJJ(COJJL()JLUGJoqcJ/JGJJJOLDGACGDLA;J2/J1JJGCL11EJ2JVTD2UCGWGfflGJJJGJJLGFigure6:Blockdiagramofthesystem’ssoftwareProc.ofSPIEVol.662566251G-6 3.3ImagedataacquisitionandtransformationThedevicesdriverisaveryimportantpartinLinuxsystem.Itestablishesstandardabstractinterfacesfordevices.Withcameradriverwecanopen/closeandreadthecameralikecommonfiles.InthisdesignweuseanOV511+tocontroltheimagesensor.TheOV511+communicateswithARMcorethroughUSBport,sothesystemshouldrecognizeUSBdeviceautomatically,andloadtheOV511+driverintokernelatstart.Wecangetinformationanddatafromcamerabycallingcorrespondingfunctionsdirectly.Nowtherawdatagottenfromcameraisnotsuitableforirisrecognition.Thewaytoresolvethisistransformingtherawirisimagedataintotheformatwhichisappropriatefortherecognitionalgorithm.WeadoptV4L(Video4Linux)totransformtherawimagedataintoYUVformat.V4Listhekerneldriveraboutvideodevices.Itprovidesagroupofinterfacefunctionsforapplicationprogramsaboutvideo.3.4IrisrecognitionapplicationprogramTheirisrecognitionprocesscontainsthreesteps:imagepre-processing,featureextractionandfeaturematching/decisionmaking.ThewholeirisrecognitionworkflowisshowninFigure7.3.4.1Imagepre-processingBeforefeatureextraction,someprocessingmustbedoneinordertogainabetterrecognitionresult.Ourpre-processingoperatesintwosteps.First,localizetheirisandremovethe“unuseful”parts(e.g.eyelid,pupiletc.)fromtheoriginalimage.Inthisdesignwefindtheiris’outeredgecursorilyusinghistogramanalyzingfirst,thenusingaHoughtransform[5][6]toconfirmitaccurately.Second,normalizethelocalizedirisinordertoachieveinvariancetothevaryingsizeofthepupilandtodistanceandangleoftheimageacquiringframework.Inthisdesign,weunfoldtheoriginalannulariristoarectangularblockafixedsize.IrisImagePre-ProcessingStandardFeatureExtractionIrisFeatureAndCodingCodeFeatureMatchingMatchingResultFigure7:Irisrecognitionprogramflow3.4.2FeatureextractionandcodingTheusefulinformationfromirisesiscontainedinabundanttexture.Unlikeothertexture-basedbiometricidentificationProc.ofSPIEVol.662566251G-7 [7]technologies,itisdifficulttoclassifyandlocalizesemanticallymeaningfulfeaturesinanirisimage.Weusea2-DGaborfiltertoextractfeaturefromirisimage.GaborfilterwasdefinedbyD.Gaborin1946.J.Daugmanexpandeditinto2-Dformin1980andfurtherelaborateditscharacterin1985.J.Daugmanpointedoutthatusing2-DGaborfiltercouldgetfrequencyanddirectionselectivitysimultaneouslywhenachievedspacelocalized.Suchcharactersareveryavailablefortextureanalysis.Wesegmentirisimageintoseveralblocksandacquirefrequencyanddirectioninformationofeach[7]sub-blockusing2-DGaborfilter.Thencodethefeatureinformationdependingonsuchinformation.3.4.3FeatureMatchingWematchobjectirisimageandstandardirisimagebyjudgingtheHammingdistancebetweentheirfeaturecodesinthisdesign.HammingdistanceisthenumberofdifferentvaluesoncorrespondingDsoftwocodes.ThelargertheHammingdistancebetweentwocodes,themoredifferenceofthosetwocodes.WesetathresholdvalueabouttheHammingdistance.IfthevalueofHammingdistanceissmallerthanthisthresholdvalue,thetwoirisimageshouldcomefromasameperson,onthecontrary,theycomefromthedifferentperson.4EXPERIMENTSWeloadedthebootloader,LinuxkernelandtheimagecapturingprograminNANDFlashmemory.OurexperimentalresultisthattheS3C2410canreadimagedatafromtheUSBcamera,storetheimagedataintoNANDFlashmemoryanddisplaythecapturedimagesthroughLCDsuccessfully.Everyirisimagehasthesizeof640×480pixels.15framesirisimagesarecapturedandtransferredpersecond.Thisresultprovesthatthedatastreamcanflowsmoothlybetweenimageacquisitionmodule,ARMcoreboardandresultdisplaymodule.Figure8issomeirisimagesacquiredbyoursystem.Figure8:IrisimagesacquiredbyoursystemWetestthealgorithm’sspeedaswell.ThemainfrequencyofS3C2410is266MHz.Itsprocessingspeedisupto290MIPS.TaketheembeddedLinuxsystemandI/Ocontrollingintoaccount,theprocessingabilityforirisrecognitionwillbeabout200MIPS.Thisisfastenoughforreal-timeirisimageacquisitionandrecognition.Throughtransplantationandoptimization,wecanshortenthisprocessingtime.Sothissystemhasthecapacitytoprocesstheirisimagedataandgivetherecognitionresultinanacceptabletime.5CONCLUSIONIt’saprogressofrealizingembeddedirisrecognitionsystem.Wedescribeourdesignmethodinthispaper.Basedonthesemethods,wedesigntheexperimentalprototypeoftheembeddedrecognitionsystemwithS3C2410.ThissystemisProc.ofSPIEVol.662566251G-8 smaller,lighterandwithlowerpowerconsumption,soitismoreconvenientthanthePC-basedirisrecognitionsystem.Thanksfortheusingofembeddedsystem,wecantransplantalgorithmsconvenientlyfromPC.Becauseoftheopensourcecode,itisfreertodosoftwaredevelopmentonLinux.Experimentalresultsshowthatit’saneffectivemethodofusingARMtoactualizeembeddedirisrecognitionsystem.ACKNOWLEDGEMENTThisprojectissupportedbyNationalScienceFoundationofChina(No.60572058)andExcellentYoungScholarsResearchFundofBeijingInstituteofTechnology(No.2006Y0104).REFERENCES1DavidZhang,JosephP.Campbell,DavideMaltoni,RuudM.Bolle,“GuestEditorialSpecialIssueonBiometricSystems”,IEEETransactionsonSystems,Man,andCybernetics-PartC:ApplicationsandReviews,VOL.35,273~275,20052GeraldO.Williams,“Irisrecognitiontechnology”,IEEEAESSystemsMagazine,23~29,19973ChenYing,WangJunhua,LuHengli“DesignofIrisRecognitionSystemBasedonDSP”,Micro-ComputerInformation,VOL.22,149~150,20064YuqingHe,JialiCui,TieniuTan,YangshengWang,“Keytechniquesandmethodsforimagingirisinfocus”,The18thInternationalConferenceonPatternRecognition,20065XingMing,ZhihuiLi,YuanningLiu,“IrisImageAcquirementandPreprocessinginBiometrics”,proc.ofSPIE,VOL.6024,602414-2~602414-7,20056HanhoSung,JackyungLim,Ji-hyunPark,YillbyungLee,“IrisRecognitionUsingCollaretteBoundaryLocalization”,Proceedingsofthe17thInternationalConferenceonPatternRecognition(ICPR’04)7CAOJiangzhong,DAIQingyun,HEJiafeng,FANGJi,“AlgorithmforIrisRecognitionBasedonTextureDistributionandGaborFilter”,ComputerEngineering,VOL.32,199~205,2006Proc.ofSPIEVol.662566251G-9
