视频序列中的行为识别研究进展_徐勤军

28420144JOUＲNALOFELECTＲONICMEASUＲEMENTANDINSTＲUMENTATIONVol.28No.4·343·2013-07ＲeceivedDate2013-07*2013M531504DOI10．13382/j．jemi．2014．04．001*徐勤军吴镇扬210096、、、。、。TN911TP391A520．2040ＲesearchprogressonactivityrecognitioninvideoXuQinjunWuZhenyangSchoolofInformationScienceandEngineeringSoutheastUniversityNanjing210096AbstractHumanactivityrecognitioninvideosequencesisaveryactiveresearchfieldofcomputervisionwithbroadapplicationprospectsinvisualsurveillancehuman-computerinteractionmedicalcarevideoretrievalandetc．Inthispaperacomprehensiveanalysisofrecentdevelopmentsofactivityrecognitionisgiven．Firstlythedif-ferentcharacteristicsofcurrentmainstreamdatabasesareanalysed．Thenthemethodsofthefeaturerepresentationandthelatestprogressoftherecognitionalgorithmsaresummarized．Especiallyastherealapplicationsrequireal-gorithmcapableoffastatomaticandreal-timeprocessinganditisdifficulttolabelalotofvideosmanualyunsu-pervisedlearningearnsmoreandmoreattention．Finallysomedetaileddiscussionsonresearchchallengesandfu-turedirectionsinactivityrecognitionarealsoprovided．Keywordsactivityrecognitioncomputervisionpatternrecognitionvideosemanticinformationmachinelearn-ing1。。、、、。·344·28。1。Weinland2。Ｒyoo3--。4socialsignalprocess。、3。22001Weizmanninstitute20055。、、。KTHWeizmann。KTH620041。4256599。、。2005WeizmanninstituteWeizmannactionsasspace-timeshapesdata-set7Weizmann。109。1。。KTHWeizmann。1。。1IＲISAHollywood、Holly-wood-28。。2009Hollywood-28Hollywood11210366969Hollywood。、、、、、。2UniversityofcentralFloridaUCF20075-9UCFsportsactiondatasetUCFYoutubeactiondatabaseUCF50UCF101。BBC/ESPN、YouTube。UCF101、13320101。3Stanforduniversity2010OlympicsportsdatasetUCFsportsactiondataset10。YouTube1650。4UT-interactionda-tabase116shaking-handspointinghuggingpushingkickingandpunching201minCali-forniaVideoWebdatabase122010non-verbalcommunication44·345·82．5min。5BrownuniversityHMDB51132011YouTube。684951101。2011WeizmanninstituteOpenUniversityofIsraelactionsimilaritylabelingASLAN144003697。。。。。KTH。。3。、、。。。。15。3．1Bobick16MEImotion-energyimagesMHImotion-historyimages。171。11Poisson。Blank71Poisson。、、、。。KTH、Weizmann、。。、。。。1Harris3DLaptev182003HarrisCuboidGaborDollar192005HessianWillems200820Dense21。2CuboidHOG/HOFHOG3DESUＲFextendedSUＲF3DSIFT21。Wang21。densesampling2005221。·346·28。3．2。Fathi23AdaBoost。151。。。Ｒaptis24。。Jain251spatio-temporalpatch1、1、。UCF50Olympics。26attributeandparts。271attrib-utes。。StanfordAndrewNg28。。stackingandconvo-lutionHollywood2YouTube53．3%、75．8%。ObjectBankSadanand291ActionBankActionBankSVM。。。11。4。。、。、。4．1。Weiz-4·347·mann。。Blank7。6SVMKTH。LiuSVMlatent-svm27。boosting30Adaboost、KTH91．8%。。。HMMhiddenmarkovmodels、MＲFmarkovianrandomfields、CＲFconditionalrandomfieldsSCFGstochasticcontext-freegrammars。31HMM11HMM。323HMM3。HMM。33CＲFconditionalrandomfield、HMMMarkovMEMMCＲF。1。Brand34。Ｒyoo35SCFGssto-chasticcontext-freegrammars。。4．2。。Wang36。37CLDScascadeoflineardynamicalsystemsmodel。O'Hara38Ward。。12。3。Dirichlet。Li22LDAlatentdirichletallocation。Niebles39LDApLSAprobabilisticlatentsemanticanalysisKTHWeizmann83．33%、90%。Wang401·348·28。。Gong411。pLSA42pLSApLSApLSA。pLSA。Zhang43SpLSAstructuralprobabilisticlatentsemanticanalysis。。SpLSApLSA。pLSALDA-HMM-LDApolylingual。1。5。。、。、、。。21。。。441、。。BoWbagofwords45-46。BoW2003Sivic45“bagofvis-ualwords”。。。Ｒyoo471BoW。。Bettadapura48。BoWLiFei-Fei101。49max-margin。BoW1。4·349·1。。。Ｒyoo471。。1。150。。1。51。6、。1。、、。1．J．2009143377-387．GUJDINGXWANGS．ASurveyofactivityanalysisalgorithmsJ．Journalofimageandgraphics2009143377-387．2WEINLANDDＲONFAＲDＲBOYEＲE．Asurveyofvision-basedmethodsforactionrepresentationsegmen-tationandrecognitionJ．ComputerVisionandImageUnderstanding20111152224-241．3AGGAＲWALJKＲYOOMS．Humanactivityanaly-sisareviewJ．ACMComputSurv20114331-43．4CＲISTANIMＲAGHAVENDＲAＲDELBUEAetal．Humanbehavioranalysisinvideosurveillanceaso-cialsignalprocessingperspectiveJ．Neurocomput-ing201310086-97．5CHAQUETJMCAＲMONAEJNDEZ-CABALLEＲOFA．AsurveyofvideodatasetsforhumanactionandactivityrecognitionJ．ComputerVisionandImageUnderstanding20131176633-659．6SCHULDTCLAPTEVICAPUTOB．ＲecognizinghumanactionsalocalSVMapproachC．2004Pro-ceedingsofthe17thInternationalConferenceonPatternＲecognition23-26Aug．200432-36．7BLANKMGOＲELICKLSHECHTMANEetal．Actionsasspace-timeshapesC．ComputerVision2005TenthIEEEInternationalConferenceon17-21Oct．20051395-1402．8MAＲSZALEKMLAPTEVISCHMIDC．ActionsincontextC．2009IEEEConferenceonComputerVi-sionandPatternＲecognition20-25June20092929-2936．9SOOMＲOKZAMIＲAＲSHAHM．UCF101adatasetof101humanactionsclassesfromvideosinthewildＲ．CＲCV-TＲ-12-01．UniversityofCentralFlorida2012．10NIEBLESJCHENCWFEI-FEIL．Modelingtempo-ralstructureofdecomposablemotionsegmentsforactivi-tyclassificationC．EuropeanConferenceonComput-erVisionECCV2012392-405．11ＲYOOMSAGGAＲWALJK．Spatio-temporalrela-tionshipmatchvideostructurecomparisonforrecogni-tionofcomplexhumanactivitiesC．2009IEEE12thInternationalConferenceonComputerVisionSept．2920091593-1600．·350·2812DENINAGBHANUBNGUYENHTetal．Vide-owebdatasetformulti-cameraactivitiesandnon-verbalcommunicationM．SpringerDistributedVideoSen-sorNetworks2011335-347．13KUEHNEHJHUANGHGAＲＲOTEEetal．HMDBal