遮挡情况下目标跟踪算法综述_薛陈

2　5200910　　　　　　　ChineseJournalofOpticsandAppliedOptics　　　　Vol.2　No.5　Oct.2009　　:2009-06-12;:2009-08-16　1674-2915(2009)05-0388-06薛　陈1,2,朱　明1,刘春香3(1.中国科学院长春光学精密机械与物理研究所,吉林长春130033;2.中国科学院研究生院,北京100039;3.长春理工大学电子信息工程学院,吉林长春130022):。,。,、、、,。,,。,。　　:遮挡跟踪;中心加权;子块匹配;轨迹预测;贝叶斯理论;多算法融合:TP301.6;TP391　　:AReviewoftrackingalgorithmsunderocclusionsXUEChen1,2,ZHUMing1,LIUChun-xiang3(1.ChangchunInstituteofOptics,FineMechanicsandPhysics,ChineseAcademyofSciences,Changchun130033,China;2.GraduateUniversityofChineseAcademyofSciences,Beijing100039,China;3.SchoolofElectronicandInformationEngineering,ChangchunUniversityofScienceandTechnology,Changchun130022,China)Abstract:Theocclusionsoccurredinthetargettrackingisoneofthedifficultiesinimageprocessing,whichhasbecomeacrucialfactoroftherobustnessoftrackingalgorithmtodealwiththeocclasionseffectivelyintracking,especallysevereocclusionandtotalocclusion.Thispapersummarizesandclassifiesthemaintrack-ingalgorithmsunderocclusionscommonlyusedathomeandabroad,andintroducesseveralkindsofpopulartrackingalgorithmsunderocclusionsbasedoncentralweights,partmatching,trackprediction,andBayesiantheory,respectively.Especially,itdescribesafusionalgorithmcombineddifferentadvantagesofsomealgo-rithms.Italsogivesthebasicideasofthesealgorithmsandtheirprocessingabilities,andanalyzestheadvan-tagesanddisadvantagesofthesealgorithms.Finally,itfocusontheproblemsandpointsoutthatthemulti-sensorfusiontechnologywillbeconductivetotheimprovementofthetrackingalgorithmsunderocclusions.Keywords:trackingunderocclusion;centralweight;partmatching;trackprediction;Bayesiantheory;fusionofmulti-algorithms1　引　言　　,、、,、、、、。,,。,、、,。。,,;,,;,。:,;,;,。,,,。,,。,,,,。,,,。:1);2);3);4);5)。,,,。2　遮挡目标跟踪算法2.1　,,。,Mean-shift[1]。Mean-shift,。,,,,。,,,。,,。,[2,3],,,,,。2.2　,,,。,,。,[7～14]。,。,。,3895　　　　　　　,:;,。2.3　,、、。,,,,。,。。[15],,[16]。,,,,。,。,(ExtendedKalmanFilter)[17](UnscentedKalmanFilter)[18]。2.4　、,。,,[19,20]。(Particlefilter)[21],Mean-shift。()。,。[22],,。,,,,。、,。2.5　,,,。[23]3,α-β,,。[24]Mean-shift,,,。、、,。[25]Mean-shift、SIFT[26,27]Kalman。Mean-shift,,SIFT,,Kalman。3　遮挡算法性能分析　　:;,,,。,,,。,,,。,、;、,。,。,:1):≤10frame;10frame。2):≤40%;40%;。3):≥15frame/s。390　　　　　　　　　　2　。,、、、、。15。1　Tab.1　Representativepapersofdifferentalgorithmsandchoseninformation2.1[1]＊N/AN/AN/A2.2[8]＊N/AN/AN/A2.3[16]＊＊N/AN/A2.4[22]＊＊＊N/A2.5[24]＊＊N/A＊,;,。,5,2。2　Tab.2　Performanceofmentionedalgorithmstoocclusions/frame2.12984150frame/s2.222050frame/s2.313N/A10N/A2.41025frame/s425frame/s2.522N/A11N/A　4　遮挡算法优缺点分析　　,,。Mean-shift,,,。Mean-shift,,,。,,。,Mean-shift[4～6]。,,。,。,,;,,,。,,,,。,。,,、,。,,、,。、,。,,,。、、、,,,。3915　　　　　　　,:,,。,。,,;,,。5　结束语　　,,,。:1.。、,、。,,,,,,。,,,,,,。,。2.。,,,。,,。,。3.,。。,。,,,,。,。,,,,。,、、。:[1]　COMANICIUD,RAMESHV,MEERP.Kernel-basedobjecttracking[J].IEEET.PatternAnal.,2003:564-577.[2]　江泽涛,赵榕春,黎明.[J].航空学报,2006,27(4):670-675.JIANGZT,ZHAORCH,LIM.Acorrelationbasedlayeredmatchingandtargettrackingmethod[J].ActaAeronauticaetAstronauticsSinica,2006,27(4):670-675.(inChinese)[3]　王明波,周亚凡,刘颖.[J].现代雷达,2008,30(7):52-55.WANGMB,ZHOUYF,LIUY.Astudyontrackingmethodformovingtargetunderscreening[J].ModernRadar,2008,30(7):52-55(inChinese).[4]　JEYAKARJ,VENKATESHBR,RAMAKRISHNANKR.Robustobjecttrackingusinglocalkernelsandbackgroundin-formation[C].IEEEInt.Conf.onImageProc.,2007:49-52.[5]　JEYAKARJ,VENKATESHBR,RAMAKRISHNANKR.Robustobjecttrackingwithbackground-weightedlocalkernels[J].Comput.Vis.ImageUnd.,2008:1-14.[6]　WENZHQ,CAIZX.ArobustobjectrackingapproachusingMeanShift[C].3thInt.Conf.onNat.Computat.(ICNC2007),2007.392　　　　　　　　　　2　[7]　MAGGIOE,CAVALLAROA.Multi-parttargetrepresentationforcolortracking[C].ICIP2005.IEEEInt.Conf.onIm-ageProc.,2005,1:729-732.[8]　WANGFL,YUSHY,YANGJ.Anovelfragments-basedtrackingalgorithmusingMeanShift[C].200810thIntl.Conf.onControl,Automation,RoboticsandVision,Hanoi,Vietnam,2008:694-698.[9]　CAULFIELDD,DAWSON-HOWEK.Evaluationofmulti-partmodelsforMean-Shifttracking[C].Int.MachineVisionandImageProc.Conf.,2008,77-82.[10]　ZHANGZ,GUNESH,PICCARDIM.Trackingpeopleincrowdsbyapartmatchingapproach[C].IEEE5thInt.Conf.onAdvancedVideoandSignalBasedSurveillance,2008:88-93.[11]　孙中森,孙俊喜,宋建中.[J].电子应用技术,2007,3:55-57.SUNZHS,SUNJX,SONGJZH.Acolortargettrackingalgorithmbasedonmulti-partrepresentation[J].Appl.Elec-tronicTechnique,2007,3:55-57.(inChinese).[12]　常发亮,马丽,乔谊正.[J].中国图像图形学报,2006,11(6):877-882.CHANGFL,MAL,QIAOYZH.Targettrackingalgorithmunderocclusionbasedoncorrelationmatching[J].J.ImageandGraphics,2006,11(6):877-882.(inChinese)[13]　常发亮,马丽,乔谊正.[J].控制与决策,2006,21(5):503-507.CHANGFL,MAL,QIAOYZH.Studyonvisiontargettrackingunderocclusion[J].ControlandDecision,2006,21(5):503-507.(inChinese)[14]　ADAMA,RIVLINE,SHIMSHONII.Robustfragments-basedtracingusingintegralhistogram[C].Proc.oftheIEEESocietyConf.onComput.Vis.andPatternRecognition,2006:798-805.[15]　KALMANRE.Anewapproachtolinearfilteringandpredictionproblems[J].Trans.ASME-J.ofBasicEng.,1960,82(SeriesD):35-45.[16]　WANGZHQ,FANYF,ZHANGGL,etal..Robustfacetrackingalgorithmwithocclusions[J].SPIE,2007,6786:67861X.[17]　RIBEIROMI.KalmanandextendedKalmanfilter:concept,derivati