資源描述:
《基于mean+shift算法的運(yùn)動(dòng)目標(biāo)檢測(cè)與跟蹤》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、摘要摘要視頻圖像序列中運(yùn)動(dòng)目標(biāo)檢測(cè)與跟蹤技術(shù),是計(jì)算機(jī)視覺(jué)領(lǐng)域的一個(gè)重要分支,它融合了圖像處理、模式識(shí)別、以及計(jì)算機(jī)等若干領(lǐng)域的先進(jìn)技術(shù),在視頻監(jiān)控、智能交通、機(jī)器人視覺(jué)導(dǎo)航以及醫(yī)療診斷等方面有著廣泛的應(yīng)用前景。本文主要圍繞運(yùn)動(dòng)目標(biāo)檢測(cè)問(wèn)題和運(yùn)動(dòng)目標(biāo)跟蹤問(wèn)題展開:對(duì)于運(yùn)動(dòng)目標(biāo)檢測(cè),首先研究了基于像素灰度歸類的背景重構(gòu)算法,并提出了背景漸變和突變時(shí)背景更新策略,然后研究了基于RGB和HSV顏色空間的移動(dòng)陰影檢測(cè)與消除方法。對(duì)于運(yùn)動(dòng)目標(biāo)跟蹤,主要研究了基于MeallShiR算法的運(yùn)動(dòng)目標(biāo)跟蹤算法,并給出了基于邊緣檢測(cè)的跟蹤窗口尺度自動(dòng)更新的
2、MeallShiR跟蹤算法,以解決MeanShiR算法跟蹤窗口固定,不能跟蹤漸近或漸遠(yuǎn)的運(yùn)動(dòng)目標(biāo)的缺點(diǎn);針對(duì)MeaIlShiR算法不能較好地跟蹤被大面積遮擋或完全遮擋的運(yùn)動(dòng)目標(biāo)的缺點(diǎn),引入了粒子濾波算法及其在運(yùn)動(dòng)目標(biāo)跟蹤中的應(yīng)用,并提出了多樣性粒子重采樣方法以解決粒子濾波算法的粒子貧化問(wèn)題;針對(duì)粒子濾波跟蹤算法需要的粒子量大,運(yùn)算速度慢的缺點(diǎn),本文給出了基于MeaIlShiR算法的粒子濾波算法,該算法利用MeaIlShiR算法的聚類功能將粒子聚集在更毗鄰目標(biāo)真實(shí)位置的區(qū)域,使得每個(gè)粒子更合理地表達(dá)目標(biāo)的真實(shí)狀態(tài),故算法對(duì)粒子數(shù)量的要求減
3、小,算法的運(yùn)算速度得到了提高。實(shí)驗(yàn)表明,新算法的跟蹤效果優(yōu)于傳統(tǒng)的基于MeanSlliR算法的運(yùn)動(dòng)目標(biāo)跟蹤算法,實(shí)時(shí)性優(yōu)于單純的粒子濾波跟蹤算法。關(guān)鍵詞運(yùn)動(dòng)目標(biāo)檢測(cè),運(yùn)動(dòng)目標(biāo)跟蹤,均值漂移,粒子濾波AbstractMoVingtargetdetectionandtrackingtechnologyinVideoimagesequencesisanimportaIltbranchofcomputervision6eld,whichme略esimageprocessing,pattemreco印ition,comput%aIldothcra
4、dVancedtechnolo百es.ItiswidelyappliedtoVide0surveillance,intelligellttransportation,robotVisualnaVigation,alldmedicaldia盟osis.Basedontheprobl鋤ofmoVingtargetdetectionandtracking,themainworkofthisdissenationcallbepresentedasfollows:ForthemoVingta昭etdetection,aback伊oundrecon
5、stnlctionalgoritllIIlbasedonpixelintensityclaSsificationisresearched,andabackgroundupdatemethodispresentedbaLsedontheback黟ound伊adualchangea11dHmtation.Asforshadowdetection趾dshadowr鋤oval,analgorithminthebasisofRGBandHSVcolorspaceisintroduced.ForthemoVingta唱ettracl(in呂anob
6、jecttrackingalgoritIllIlbasedonMeaJlShiftalgoritllIIlisresearched,manyimprov鋤entsaredoneforthedefectsofthetrackingalgo“thm.AMeanShiRtrackingalgoritIInlwithself二updatingtrackingwindowbaSedonobjectsedgedetectionwaspresentedforMeallShiRtracl【ingalgorithm’slackingofbandwidth
7、adaptationmechanism;111ordertosolVemedefectofMeallShiRobjecttr.a(chǎn)ckingalgorithminmo“ngobject仃ackingunderocclusion,P硪icleFilteralgoritllrrlaIlditsapplicationinmoVingtarget仃ackingisintrDduced,鋤dadiVersityp砒icleres鋤plemeⅡlodisproposedinofdert0resolVetheprobl鋤ofpanicleimpoVer
8、isllIllent.T11emaindefectofPanicleFilteristheneedsofmlmerouspaniclestoestimatemestateofobjects,w11ichin