资源描述：
《1999-IEEE-trans-An HMM-based threshold model approach for gesture recognition》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。
1、IEEETRANSACTIONSONPATTERNANALYSISANDMACHINEINTELLIGENCE,VOL.21,NO.10,OCTOBER1999961AnHMM-BasedThresholdModelApproachforGestureRecognitionHyeon-KyuLeeandJinH.KimAbstractÐThetaskofautomaticgesturerecognitionishighlychallengingduetothepresenceofunpredictableandambiguousnongesturehandmotions.Inth
2、ispaper,anewmethodisdevelopedusingtheHiddenMarkovModelbasedtechnique.Tohandlenongesturepatterns,weintroducetheconceptofathresholdmodelthatcalculatesthelikelihoodthresholdofaninputpatternandprovidesaconfirmationmechanismfortheprovisionallymatchedgesturepatterns.Thethresholdmodelisaweakmodelfor
3、alltrainedgesturesinthesensethatitslikelihoodissmallerthanthatofthededicatedgesturemodelforagivengesture.Consequently,thelikelihoodcanbeusedasanadaptivethresholdforselectingpropergesturemodel.Ithas,however,alargenumberofstatesandneedstobereducedbecausethethresholdmodelisconstructedbycollectin
4、gthestatesofallgesturemodelsinthesystem.Toovercomethisproblem,thestateswithsimilarprobabilitydistributionsaremerged,utilizingtherelativeentropymeasure.Experimentalresultsshowthattheproposedmethodcansuccessfullyextracttrainedgesturesfromcontinuoushandmotionwith93.14percentreliability.IndexTerm
5、sÐHandgesture,gesturespotting,HiddenMarkovModel,segmentation,patternrecognition,relativeentropy,statereduction,thresholdmodel.æ1INTRODUCTIONHUMANgesturesconstituteaspaceofmotionexpressedcomesfromthefactthatthesamegesturevariesdynami-bythebody,face,and/orhands.Amongavarietyofcallyinshapeanddur
6、ation,evenforthesamegesturer.Angestures,thehandgestureisthemostexpressiveandtheidealrecognizerwillextractgesturesegmentsfromthemostfrequentlyusedone[1],[2],[3],[4],[5].Inthispaper,continuousinputsignalandmatchthemwithreferencewedefineagestureasameaningfulpartofthehandmotionpatternsallowingawi
7、derangeofspatio-temporaltocommunicatewithacomputer.Theinteractionusingvariability.handgestureshasbeenstudiedasanalternativeformofRecently,theHMMhasattractedtheattentionofmanyhuman-computerinterfacebyanumberofresearchers,researchersasausefulto
