siamese neural network based gait recognition for human ... · network based gait recognition for...

25
Siamese Neural Network based Gait Recognition for Human Identification Cheng Zhang, Wu Liu, Huadong Ma, Huiyuan Fu Beijing University of Posts and Telecommunications ICASSP 2016

Upload: others

Post on 11-Jul-2020

9 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

SiameseNeuralNetworkbasedGaitRecognitionforHumanIdentification

ChengZhang,WuLiu,Huadong Ma,Huiyuan FuBeijingUniversityofPostsandTelecommunications

ICASSP2016

Page 2: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

Outline

• Introduction• Proposedmethod– ConventionalCNNbasedGaitRecognition– SiameseNetworkbasedGaitRecognition

• Experiments• Conclusions

Page 3: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

Definition• Gait analysis is the systematic study of animal locomotion,

more specifically the study of human motion, using the eyeand the brain of observers, augmented by instrumentation formeasuring body movements, body mechanics, and theactivity of the muscles.

Page 4: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

Background

• Socialsecurity– Videobigdataandcameranetwork– Remotesurveillance– Identificationandattributeclassification

• Biometricauthenticationtechniques– Facialrecognition– Irisrecognition– Fingerprinttechnologies– Voiceverification– Handgeometry

Page 5: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

Characteristics

• Remoteaccessed– Itcanidentifysubjectsfromadistancewithoutinterruptingthesubject

• Robust– Eveninlowresolutionvideos,thegaitstillworkswell

• Security– Itisdifficulttoimitateorcamouflagehumangait

iris fingerprintface voice gait

Page 6: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

Whygaitworks?

• A plethora of technique and data continue to showthat a person’s walking is indeed unique

MurrayMP,DroughtAB,KoryRC.Walkingpatternsofnormalmen[J].TheJournalofBone&JointSurgery,1964,46(2):335-360.JohanssonG.Visualperceptionofbiologicalmotionandamodelforitsanalysis[J].Attention,Perception,&Psychophysics,1973,14(2):201-211.

Page 7: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

Challenges• Inconspicuousinter-classdifferencefromthedifferentpeople

• Thelargeintra-classvariationsfromthesameperson– Walkingspeeds– Viewpoints– Clothing– Belongings– Occasion normal clothes backpack

Gaitsilhouettesofdifferentsubject

Page 8: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

RecentEffortsandMajorDrawback• Model-basedmethods

– Extractinghumanbodystructurefromtheimages– Requiringahighresolutionaswellashighercomputationalcostand

arenotyetsuitableforoutdoorsurveillance

• Model-freemethods– Usingthewholemotionpattern/featuresofthehumanbody,and

performingrecognitionatlowerresolutions– Human-craftedgaitfeaturescanextremelyhardtobreakthrough

featurerepresentationbottleneckwhenfacingwiththegaitandappearancechanges

Page 9: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

GeneralStepsofOurSystem

Page 10: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

GaitEnergyImage

• Averagingofsilhouetteoveronegaitcycle– Representahumanmotionsequenceinasingleimagewhilepreservingtemporalinformation

– Robusttoincidentalsilhouetteerrorsinindividualimage

Page 11: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

ConventionalCNNbasedGR

• RetraintheCNNsonthegaitdataset– CNNsareabletolearndiscriminativefeatures– Fine-tuningfromapre-trainedmodel(e.g.,AlexNet)isagoodsolutiontosolvethedatalimitationproblemandspeeduptheconvergenceofnewmodel

– EmploytheAlexNet andonlychangethe1,000labeloutputtothenumberofsubjectsingaitdataset

Page 12: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

Problems

• Datalimitation– Tolearnsufficientfeatures,theCNNrequiresamassoftrainingdataforallcategories

– Forgaitrecognition,thenumberofsubjectscanbelarge,whilewithonlyafewexamplespersubjectinpublicdatabase

• Domaingap– Gaitrecognitionforhumanidentificationisessentiallyasearchproblembutnotclassification

Page 13: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

MetricLearning

= ≠

Page 14: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

ProposedFramework

• SiameseNeuralNetworkbasedgaitrecognition

Page 15: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

Sampling

• Trainingdataishighlyunbalanced– Usingasamplertogenerateequalnumberofpositiveandnegativeineachmini-batch,avoidoverlybiased towardstonegativedecisions

– Usingasamplertoenforcevarietytopreventoverfitting toalimitednegativeset

• Specially,thetrainingsetisselectedfromOULP-C1V1-A-Gallery dataset,with20,000similarGEIpairsandrandomlyselected20,000dissimilarpairs

Page 16: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

LossFunction

• ThedistancebetweenapairofGEIscanbemeasuredby:

• Wecandefinethecontrastivefunctionasfollows:

Page 17: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

TrainingandFeatureExtraction

• Supervisedsetting• MinimizedthecontrastivelossfunctionoveratrainingsetofNpatchpairsusingstochasticgradientdescent

• Experimentedwithdifferentparametersandgavethebestperformanceoffeaturerepresentation

Page 18: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

Experiments• Database:OU-ISIRLargePopulation

• Evaluation:Rank-1andRank-5identificationrates

• Baselines:STOAgaitrecognitionmethods,i.e.,GEI,FDF,HWLD,VTM,andRankSVM

• Pipeline:Backgroundsegmentation->Periodicidentification->GEIsgeneration->DNNtraining->DNNfeatureextraction->K-Nearest-Neighborsearching

Page 19: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

Database• OU-ISIRLargePopulationGaitDatabase

– Containstheworld’slargestnumberofsubjects– Recordstwosequencesforeachsubject:probe(query)andgallery

(source)sequence,offersfaircomparisontestbed

Page 20: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

Intra-viewrecognition

[9]H.Iwama,M.Okumura,Y.Makihara,andY.Yagi,“Theou-isirgaitdatabasecomprisingthelargepopulationdatasetandperformanceevaluationofgaitrecognition,”IEEETIFS.[7]Sivapalan,D.Chen,S.Denman,S.Sridharan,andC.Fookes,“Histogramofweightedlocaldirectionsforgaitrecognition,”inCVPRW,2013.

Page 21: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

SomeResults

Page 22: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

Inter-viewrecognition

D.Muramatsu,A.Shiraishi,Y.Makihara,M.Uddin,andY.Yagi,“Gait-basedpersonrecognitionusingarbitraryviewtransformationmodel,”IEEETIP,2015.R.Mart´ın-F´elez andT.Xiang,“Gaitrecognitionbyranking,”inECCV,2012.

Page 23: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

Conclusions• Wepresentoneofthefirstattemptstostudythedeepneural

networkbasedgaitrecognitionforhumanidentificationwithdistancemetriclearning

• Intheend-to-endframework,weleveragethecompetitiveGEIpresentationastheinputofnetworkwhileholisticallyexploittheSiameseneuralnetworktolearneffectivefeaturerepresentationsforhumanidentification

• Thecomprehensiveevaluationsshowthatweimpressivelyoutperformthestate-of-the-artsontheworld’slargestchallengegaitbenchmarkdataset

Page 24: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

FutureWorks

• 3-DimensionalSiameseneuralnetwork• Quasi-periodicorsub-framegaitrecognition• Unconstrainedenvironment,likeilluminationchanges,darkillumination,clutteredbackground,motionblur,andimagecompressionnoise

• …

“High’st Queenofstate,GreatJunocomes;Iknowherbyhergait”—— TheTempest[Act4Scene1],Shakespeare

Page 25: Siamese Neural Network based Gait Recognition for Human ... · network based gait recognition for human identification with distance metric learning • In the end-to-end framework,

Anyquestions?