用于行人头部特征提取的目标区域匹配方法

14320093JournalofImageandGraphicsVol.14,No.3Mar.,20091)1)2)1)(,310018)2)(,310027),,,,;,;,,,,,90%:TP391141:A:100628961(2009)0320482207AnTargetRegionMatchingAlgorithmfortheHeadFeatureExtractionofPedestrianYUHai2bin1),LIUJing2biao1),LIUJi2lin2)1)(CollegeofElectronicandInformation,HangzhouDianziUniversity,Hangzhou310018)2)(DepartmentofInformationandElectronicEngineering,ZhejiangUniversity,Hangzhou310027)AbstractInordertoacquireaccuratepassengerflowinformationbylocatingandtrackingmovingpedestriansaccuratelyinimagesequence,anovelapproachofheadfeatureextractionbasedontargetregionmatchingispresented1Deferentfromthecommonmethodsbasedondensedisparityimagetoobtainthedisparityoftheheadregion,themethodinthispaperisbasedontheideaofsegmentationbeforematching,i1e1thereferenceimageissegmentedfirstlybymonocularimageprocessingandthecandidateheadregionswillbeacquired.Thenthesecandidateheadregionswillbetakenasthetargetregionstobedirectlyusedinthecorrespondenceregionssearchingandmatchingtoobtaintheheaddisparity.Finallythedisparityofthecandidateheadregionsisusedtoextractthedepthandperspectivefeatureofthecandidateheadregionstoremovethefalseheadregions1Theperformancetestandexperimentresultsshowthatthemethodproposedinthispaperhastheadvantageofhigherprecisionofdisparityextractionandbetterreal2timeperformanceaswellasthegreatimportancetoextractthepedestrianheadpsfeaturewithhighdegreeofdistinctiontoeffectivelyeliminatethefalseheadregionssothattheaccuracyofthepassengerflowdetectioncanreachover90%1Keywordsstereo,imageprocessing,blockmatching,targetregionmatching:(863)(2006AA09Z228):2008206202;:2008210207:(1979),,20079E2mail:[e-mail]1,,,(),(,),,,2D[124]3D[528],2D[1][223][4];3D[5][628]2D,,3D,,3D,,,[5][7][8],,,,,,,3D,3D,[3]Hough;,;,2(),[3],Hough,,1,[3]()1,2,,()(2),[3],[3]3843:,33D(),()[9],,,,,,,(SADSSD)(NCC),SADMpSADMpSAD(h)=(u,v)Rp|Ir(u,v)-Im(u+h,v)|(1),Ir,Im,Rp,h,,,,,,,,3,xp1,xp2,dp=xp2-xp1;xt1,xt2,dt3Fig13Theprocessoftheblockmatchingpromotingtothetargetregionmatching,(1)RpRt,SADMtSADMtSAD(h)=(u,v)RtIr(u,v)-Im(u+h,v)(2),,,SAD(ZSAD)MtZSADMtZSAD(h)=(u,v)Rt[Ir(u,v)-Ir]-[Im(u+h,v)-Im(h)](3),IrIm(h)RtRt(h),[8]At(4),Ab,E,AbEAt,O(AbEAt),O(AtE),,,,,,484144Fig14ThesketchoftheacquisitionofthesametargetregionpsdisparityRCH(2)(3)Rt,dRCH,dRCH=arghminhe[MCHSAD/ZSAD(h)](4),edRCH5(c),5(c),dRCH[0,255]5dRCHFig15TheexampleoftheacquisitionofdRCH4,3D:(1),,H,B,f,()h,d,,h=H-Bfd(5)hmin,h=H-Bfdhmin(6)dBf(H-hmin)(7)THd=Bf(H-hmin)(7)(7),dRCHTHd,,,(2),,R,r,,,Rr=(Z-f)f=Zf(8),Z,,Z=fBd(8),Rr=Zf=fB(fd)=Bd(9)(9),dr=BR=kR(10)(10),,kR=BR,kR,RH,kR,5843:3D,[Rmin,Rmax],kRkminRkRkmaxR(11),kminR=B/Rmax,kmaxR=B/Rmin(11),kminRkmaxR,kd/r=dRCH/rCH,,kd/r(11),(11),,6,,THd=25pixel(140cm),(6(a))(16pixel20pixel)THd,6(c);6(c)(6(d)),(11),,kminRkmaxR11262110[3],6(g)6Fig16Theexampleofthefalseheadregioneliminationbasedonthecandidateheadregionpsdepthandperspectivefeature5PentiumIV310GHz,512M,VisualC++610,,[8]50,7,,[3],,,tptc,tt=tp+tc,,684147Fig17TheexampleoftheimageincludingtheheadregionextractionresultsusedinthealgorithmperformancetesthHdZ50tptctt=tp+tcd11Tab11Theperformancecomparisonbetweenthealgorithmofregiondisparityhistogrambasedonblockmatchingandthealgorithmoftargetregionmatchingtp(ms)tc(ms)tt(ms)d(pixel)1711305414513574316146212107%171130585126425615694112%1,2%,,:,,;(),3m,320240,15fps,4mm,1819mm8(b),2D,8(a)[2]8Fig18Theexperimentresults7843:2,,[2]2,,2D,2Tab12Thefieldtestresults/(/)[2][2]1521689/91101/10783/8784147%94143%2170437/3444/3935/3283119%94136%6,,,,,,2D(References)1BartoliniF,CappelliniV,MecocciA1Countingpeoplegettinginandoutofabusbyreal2timeimage2sequenceprocessing[J]1ImageandVisionComputing,1994,12(1):3624112ZhangXW,SextonG1Automatichumanheadlocationforpedestriancounting[A]1In:Proceedingsofthe6thInternationalConferenceonImageProcessingandItsApplications[C],Dublin,Ireland,1997:2:535254013YuHai2bin,LiuJi2lin1Avision2basedmethodtoestimatepassengerflowinbus[J]1JournalofImageandGraphics,2008,13(4):71627221[,1[J]1,2008,13(4):7162722.]4LinSF,ChenJY,ChaoHX1Estimationofnumberofpeopleincrowdedscenesusingperspectivetransformation[J]1IEEETransactionsonSystems,ManandCybernetics,2001,31(6):645265415BeymerD1Personcountingusingstereo[A]1In:ProceedingsofWorkshoponHumanMotion[C],Austin,Texas,USA,2000:127213316NarayananK,KumaranR,GowdyJ1Stereo2basedellipticalheadtracking[A]1In:ProceedingsofEuropeanSignalProcessingConference[C],Antalya,Turkey,2005:15652156817HuangXY,LiLY,SimT1Stereo2basedhumanheaddetectionfromcrowdscenes[A]1In:ProceedingsofInternationalConferenceonImageProcessing[C],Singapore,2004:13532135618LuoRJ,GuoY1Real2timestereotrackingofmultiplemovingheads[A]1In:ProceedingsofIEEEICCVWorkshoponRecognition,Analysis,andTrackingofFacesandGesturesinReal2TimeSystems[C],Vancouver,BC,Canada,2001:5526019BrownM,BurschkaD,HagerG1Advancesincomputationalstereo[J]1IEEETransactionsonPatternAnalysisandMachineIntelligence,2003,25(8):99321008188414