基于人体步态利用神经网络鉴定性别(IJMECS-V4-N11-7)

I.J.ModernEducationandComputerScience,2012,11,70-75PublishedOnlineDecember2012inMECS()DOI:10.5815/ijmecs.2012.11.07Copyright©2012MECSI.J.ModernEducationandComputerScience,2012,11,70-75GenderIdentificationinHumanGaitUsingNeuralNetworkRichaShuklaDept.ofComputerScience,SRCEM,Banmore,Morena,IndiaEmail:[e-mail],OrientalUniversity,Indore,IndiaAnupamShuklaDept.ofComputerScience,ABV-IndianInstituteofInformationTechnologyandmanagement,Gwalior,IndiaSanjeevSharmaDept.ofComputerScience,ABV-IndianInstituteofInformationTechnologyandmanagement,Gwalior,IndiaNirupamaTiwariDept.ofComputerScience,SRCEM,Banmore,Morena,IndiaAbstract—Biometricsisanadvancedwayofpersonrecognitionasitestablishesmoredirectandexplicitlinkwithhumansthanpasswords,sincebiometricsusemeasurablephysiologicalandbehaviouralfeaturesofaperson.Inthispapergenderrecognitionfromhumangaitinimagesequencehavebeensuccessfullyinvestigated.Silhouetteof15malesand15femalesfromthedatabasecollectedfromCASIRsitehavebeenextracted.Thecomputervisionbasedgenderclassificationisthencarriedoutonthebasisofstandarddeviation,centreofmassandheightfromheadtotoeusingFeedForwardBackPropagationNetworkwithTRAINLMastrainingfunctions,LEARNGDasadaptationlearningfunctionandMSEREGasperformancefunction.Experimentalresultsdemonstratethatthepresentgenderrecognitionsystemachieverecognitionperformanceof93.4%,94.6%,and94.7%with2layers/20neurons,3layers/30neuronsand4layers/30neuronsrespectively.WhentheperformancefunctionisreplacedwithSSEtherecognitionperformanceisincreasedby2%,2.4%and3%respectivelyfor2layers/20neurons,3layers/30neuronsand4layers/30neurons.TheabovestudyindicatesthatGaitbasedgenderrecognitionisoneofthebestreliablebiometrictechnologythatcanbeusedtomonitorpeoplewithouttheircooperation.Controlledenvironmentssuchasbanks,militaryinstallationsandevenairportsneedtoquicklydetectthreatsandprovidedifferinglevelsofaccesstodifferentusergroups.Indexterm—GenderRecognition,Gait,Silhouette,Featureextraction,NeuralnetworkI.INTRODUCTIONGaitrecognitionistheprocessofidentifyinganindividualbyaparticularwayormannerinwhichtheywalk.Withoutanyinteractionorco-operationfromthesubject,gaitoffersthepossibilitytoidentifypeopleatadistanceinlessunobtrusivebiometricway.Thisisthemostimportantpropertywhichmakesitsoattractive.Humangaitrecognitionasanewbiometricaimedtorecognizepersonviathestyleofpeoplewalking,whichcontainthephysiologicalorbehaviouralcharacteristicsofhuman.Gait-basedgenderclassificationisstillimmaturebecauseofitsuniqueadvantagesofbeingnoncontact,non-invasive,andeasilyacquiredatadistance,itisgainingincreasinginterestfromresearchers.HumanGaitclassificationandrecognitiongivingsomeadvantagecomparedtootherrecognitionsystem.Gaitclassificationsystemdoesnotrequireobservedsubject’sattentionandassistance.Itcanalsocapturegaitatafardistancewithoutrequiringphysicalinformationfromsubjects[1-3].Thereisasignificantdifferencebetweenhumangaitandotherbiometricsclassification.Inhumangait,weshouldusevideodatainsteadofusingimagedataasotherbiometricssystemusedwidely.Invideodata,wecanutilizespatialdataaswellastemporaldatacomparetoimagedata.Mostofthegaitclassificationandorrecognitionsystemcreatedareusingspatialdataonly[4-11].HumanGaitGenderClassificationasarecognitionsystemdividedinthreemainsubject;preprocessing,featureextractionandclassification.Inmostgaitrecognitionapproaches[12-14]recognitionfeaturesareextractedfromsilhouetteimages.DimosthenisIoannidiset.al.(2007)[15]proposedaninnovativegaitidentificationandauthenticationmethodbasedontheuseofnovel2-Dand3-Dfeatures.Thehumansilhouetteissegmentedintosevencomponents,namelyhead,arm,trunk,thigh,front-leg,back-leg,andfeet.Thelegsilhouettesforthefront-legandtheback-legareconsideredseparatelybecause,duringwalking,theleftlegandtherightlegareinfrontoratthebackbyturns.EachofthesevencomponentsandanumberofcombinationsoftheGenderIdentificationinHumanGaitUsingNeuralNetwork71Copyright©2012MECSI.J.ModernEducationandComputerScience,2012,11,70-75componentsarethenstudiedwithregardtotwousefulapplications:humanidentification(ID)recognitionandgenderrecognition.Manysocialinteractionsdependgreatlyoncorrectgenderperception.Ifacomputerbasedbiometricsystemcanrecognizegender,itwillbeveryhelpfulinmanyapplications.Forexample,genderclassificationcanimprovesurveillancesystems,intelligence,analysecustomersforstoremanagers,allowrobotstoperceivegender,etc.Automaticgenderclassificationcanbebasedonface[16-19]orgait[20].Anovelmultimodalbiometricrecognitionsystemusingthreemodalitiesincludingface,earandgait,basedonGabor+PCAfeatureextractionmethodwithfusionatmatchingscorelevelisdonebyAliPourYazdanpanahet.al.(2010)[21].Thispaperpresentsgenderidentificationusinggaitbiometricsmethodbaseduponsilhouetteanalysismeasuredduringwalking.Recognizingpeoplebygenderdependsgreatlyonhowthesilhouetteshapeofanindividualchangesovertimeinanimagesequence.So,wemayconsidergaitmotiontobecomposedofasequenceofstaticbodygestureandexpectthatsomedistinguishablesignatureswithrespectt