基于背景剪除和隱馬爾可夫模型的人體動作識別.pdf

1、安徽大學(xué)碩士學(xué)位論文基于背景剪除和隱馬爾可夫模型的人體動作識別姓名：齊登鋼申請學(xué)位級別：碩士專業(yè)：電子與通信工程指導(dǎo)教師：徐超；譚守標(biāo)201104ABSTRACTAsthedevelopofscienceandtechnologyartificialintelligenceforreconnitionhaslargelydevelopedCurrentlypeoplepaymoreandmoreattentionontheresearc

2、hofhumanmotionrecognitionHumanactionrecognitionhasawiderangeofapplicationsinintelligentsurveilancehumancomputerinteractionvirtualrealitymotionanalysisandotherfieldshasawiderangeofapplicationsBasedontheresearchofhumanmoti

3、onrecognitionhomeandabroadthispaperimprovedthetraditionalHMMrecognitionalgorithmAndthispaperpresentsanewmethodbasedonthebackgroundcutoffandEHMMcombinationofhumanactionrecognitionmethodThemaincontentsareasfollows：Firstlyt

4、heactionsequencesforvideoadoptedgaussianmixturemodelmethodfortargetdetectionSecondly，cutoffbackgroundforeachfTameofvideoimages，extracttheforegroundsilhouetteofhumanactiontofindoutbodysilhouette，preraringforthenextactionT

5、hirdlyforthefeatureextractionthebodydividedintofiveparts，containinghead，孤l璐，waistthighs，legsandfeetEachpartasasuperstatgeverysuperstatecontainingseveralpartsasembeddedstatesFourthlyC強(qiáng)11Tiesonthefeatureextractionformtheob

6、servationvectorsequenceFifthlytrainingEHMMmodel，anddepositefeaturevectorsi11t0theEHMMfeaturelibraryLastofall，tocomparefeaturrevectorsintheEHMMfeaturelibraryandthefeaturrevectorsoftheneedrecognizingactionthenfindoutthemax

7、imumprobabilitymatchingfeaturevectorandthusreoognizethehumanactionInthepilotphasethetwosetsofdifferentexperimentsweredesignedInthefirstsetofexperiments，thearticleselectedl0movementsofcommonlifeAtthesecondsetofexperiments