資源描述:
《基于角度及輪廓特征的步態(tài)識(shí)別方法分析》由會(huì)員上傳分享,免費(fèi)在線閱讀,更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫(kù)。
1、摘要隨著現(xiàn)代社會(huì)對(duì)安全的迫切需求,生物特征識(shí)別作為對(duì)安全敏感場(chǎng)合進(jìn)行訪問控制的一種身份鑒別技術(shù),得到大力的研究和發(fā)展。人臉、指紋等生物特征通常要求近距離或接觸性的感知,在遠(yuǎn)距離的情況下將無(wú)法使用,而步態(tài)則是此種情況下唯一可感知的生物特征,因此,步態(tài)識(shí)別——利用人行走的姿勢(shì)鑒別人體身份,在視覺監(jiān)控領(lǐng)域具有顯著的優(yōu)勢(shì)。首先對(duì)步態(tài)進(jìn)行預(yù)處理。通過分析和比較常用的運(yùn)動(dòng)檢測(cè)方法,針對(duì)文中使用的步態(tài)序列具有背景簡(jiǎn)單的特點(diǎn),采用背景減除法實(shí)現(xiàn)步態(tài)檢測(cè);對(duì)步態(tài)運(yùn)動(dòng)的周期性進(jìn)行分析,并利用人體側(cè)影的高寬信息計(jì)算步態(tài)周期。然后,基于“行走運(yùn)動(dòng)的關(guān)節(jié)角度變化包含著豐富的個(gè)體識(shí)別信息”的觀點(diǎn),提出了基于下肢運(yùn)動(dòng)分析
2、及DTW的步態(tài)識(shí)別方法。對(duì)每個(gè)步態(tài)序列,依據(jù)人體解剖學(xué)的先驗(yàn)知識(shí),通過對(duì)下肢運(yùn)動(dòng)的分析來定位關(guān)節(jié)點(diǎn),進(jìn)而提取一個(gè)步態(tài)周期的下肢角度變化軌跡作為特征矢量;采用動(dòng)態(tài)時(shí)間規(guī)整(DTW)算法計(jì)算不同步態(tài)序列問的相似度,使用最近鄰分類器及K近鄰分類器實(shí)現(xiàn)步態(tài)的分類決策。最后,針對(duì)基于單一特征的步態(tài)識(shí)別方法一般不能得到令人滿意的識(shí)別效果,利用特征融合的思想,提出了在匹配層上融合下肢角度及人體輪廓特征的步態(tài)識(shí)別方法。融合過程中,權(quán)重的賦予使兩種步態(tài)特征按較優(yōu)的比例進(jìn)行組合,以充分發(fā)揮其互補(bǔ)性。CASIA數(shù)據(jù)庫(kù)上的仿真結(jié)果表明,本方法的識(shí)別率達(dá)到95%,明顯高于基于單一特征的步態(tài)識(shí)別方法。關(guān)鍵詞生物特征識(shí)別
3、,步態(tài)識(shí)別,動(dòng)態(tài)時(shí)間規(guī)整,特征融合,傅立葉描述子ABSTRACTWithagrowingneedforsecurityinmodernsociety,biometricsrecognition,asahumanidentificationapproachforaCCeSScontrolinsecurity.sensitiveenvironments,hasbeengreatlyresearchedanddeveloped.Tooperatesuccessfully,theestablishedbiometricssuchasface、fingerprintusuallyrequireprox
4、imalsensingorphysicalcontact.However,Gaitistheonlyonebiometricthatcanbeeasilyperceivedatadistance.So,gaitrecognition,whichaimstoidentifyindividualsbytheirwalkingmanners,isveryattractiveinthefieldofvisualsurveillance.Atfirst,gaitsequencesarepreprocessed.Byanalyzingandcomparingkindsofmotiondetectionm
5、ethods,andconsideringthesimplebackgroundofgaitsequences,backgroundsubtractionisusedingaitdetection.Gaitcycleisanalyzed,thenwidthandheightofbodyanalysisisperformedtocomputerit.Next,basedontheideathatjoint—angletrajectoriesofbodypartsinwalkingmotionincludesufficientdynamicidentityinformation,agaitrec
6、ognitionmethodbasedonlower-limbmotionanalysisanddynamictimewrappingisproposed.Foreachgaitsequence,accordingtotheknowledgeofhumanbodyanatomy,thecoordinatesoflower-limbjointsareobtainedbyanalyzinglower-limbmotion,andthenthetrajectoriesoflower..1imbanglesinonecycleareextractedasfeaturevectors.Dynamict
7、imewrappingisusedtomeasurethesimilarityofdifferentsequences,then,nearestneighboralgorithmandK·neighboralgorithmarefinallyperformedtorealizegaitrecognition.Atlast,inordertosolvetheproblemthatmostgaitrecognit