20086223JournalofAcademyofArmoredForceEngineeringJun.2008Vol.22No.3:167221497(2008)0320059204Mean2shift1211(1.,100072;21,100072):Mean2shift,Mean2shift,,,,:;;Mean2shift:TN911.73:AAnImprovedMean2shiftAlgorithmforObjectTrackingYAOJun2jie1ZHANGGui2min2SONGAi2bin1LIUYing1(1.DepartmentofFundamentalCourses,AcademyofArmoredForceEngineering,Beijing100072,China;2.DepartmentofControlEngineering,AcademyofArmoredForceEngineering,Beijing100072,China)Abstract:Basedonthecolorhistograminformationoftargetregioninmulti2frameimages,anewMean2shiftalgorithmforobjecttrackingisputforward.FocusingonMean2shift,whenestablishingcharacteristicmodelfortargettemplate,thenewalgorithmnotonlyconsidersthecolorhistogramoftargetregionintheproceedingframe,butalsoconsidersthecolorhistogramoftargetregioninthefirstseveralframesthroughtheweighting.Therefore,thestabilityoftrackingisimproved.Keywords:histogram;objecttracking;Mean2shift2323(),,,,,,,,,;Mean2shift[1],[2],Mean2shift,[3-4][4-5],,,,2,,;,,,,,,:200800:1980-.22Mean2shift1,HSVRGB,,RGBH(Hν255),HHH,H31:Rdd,x,xx2=xTx,RK(x)RdR,k[0,+)R,K(x)=ckk(x2),(1)ck0,k:1)k;2)k,ab,k(a)k(b);3)k,k(r)drK(x),,,KN(x)=cexp-12x2(2)EpanechnikovKE=c(1-x2),x10,1(3),,xi(i=1,2,,N),,bR2{0,1,,H3-1},xi,b(xi),h=0,1,,H3-1q=N=K(x)[(x)],=,,,3,()K();()delta;C,H3-1h=0qh=1,C=1Ni=1K(xi2),,Q=(qij)H3L,qiji-1L-j+1Qqij=qi-1(j=1,2,,L;i=1,2,,H3),qLh=1LL-1j=0qh,L-j,qSh=1SS-1j=0SLqh,L-j,qh=qLh+qSh2hL,,;,,,L=S=1,xi(i=1,2,,nc),SLSW,y,h=0,1,,H3-1ph(y)=Ccnci=1K(y-xir2)[b(xi)-h],h=0,1,,H3-1,(5)K();r=S2L+S2W2;Cc,H3-1h=0ph(y)=1,Cc=1nci=1K(y-xir2),q=(q0,q1,,qH3-1)p(y)=(p0(y),p1(y),,pH3-1(y)),2,Bhatta2charyya[6]:(y)=[(y),q]=3=(y)q,(6),06hCi1i2bi-hh01H-14pH-1h0phh3:Mean2shift,2d(y)=1-[p(y),q].(7)2,(),,yy0,y0p(y0)=(p0(y0),p1(y0),,pH3-1(y0)),Bhatta2charyya[p(y),q]y0Taylor,:[p(y,)q]12H3-1h=0ph(y0)qh+12H3-1h=0ph(y)qhph(y0)1(8),2,,(5)(8)[p(y),q]12H3-1h=0ph(y0)qh+Cc2nci=1wiK(y-xir2),(9)wi=H3-1h=0�qhph(y0)[b(xi)-h].(10),,(9),(9)1y,2Mean2shift,y1[7]y1=nci=1xiwig(y0-xir2)nci=1wig(y0-xir2),(11)(x)=(x)y(y)=((y),(y),,3(y)),[p(y1),q]=H3-1h=0ph(y1)qh,[p(y1),q][p(y0),q],y1,,y1-y0Bhattacharyya,[p(y1),q][p(y0),q],y1=12(y1+y0),[p(y1),q][p(y0),q],011%,Q,qij1jL-1=qi,j+1,qi,L=p(y1)1(12)3,y0y1,10,4,Epanechnikov,y1=nci=1xiwinci=1wi,y1,,320320240Mean2shiftMean2shift,,23,12:11,8993,29395,Mean2shift(1(c)),,;Mean2shift(2(c)),,,,M2f3L,16g-k1p1p01p11pH-11eanshitH223,4Mean-shift,,,Mean2shift:[1]FukanagaK,HostetlerLD.TheEstimationoftheGradientofaDensityFunctionwithApplicationinPatternRecognition[J].IEEETransonInformationTheory,1975,21(1):32-40.[2]ComaniciuD,RameshV,MeerP.Real2timeTrackingofNon2rigidObjectsUsingMean2shift[C]ProcoftheIEEEConfonComputerVisionandPatternRecognition.HiltonHeadIsland,2000:142-149.[3],,,.Mean2shift[J].,2007,29(9):1405-1409.[4]ComaniciuD,RameshV,MeerP.Kernel2basedObjectTracking[J].IEEETransactiononPatternAnalysisandMachineIntelli2gence,2003,25(5):564-577.[5],,.Mean2shift[J].,2008,35(1):45-49.[6]ThomasK.TheDivergenceandBhattacharyyaDistanceMeasuresinSignalSelection[J].IEEETransonCommunicationTechnolo2gy,1967,15(1):52-59.[7]MarkFashing,CarloTomasi.Mean2shiftisaBoundOptimiza2tion[J].IEEETransonPattenAnalysisandIntelligence,2005,27(3):471-474.(:)26