多模型GM_CBMeMBer滤波器及航迹形成

第40卷第2期2014年2月

自动化学报

ACTAAUTOMATICASINICA

Vol.40,No.2February2014

多模型GM-CBMeMBer滤波器及航迹形成

连峰1

韩崇昭1

李晨2

摘要提出了一种可适用于杂波环境下对多个机动目标进行跟踪并能形成多目标航迹的多模型势平衡多目标多伯努利(Cardinalitybalancedmulti-targetmulti-Bernoulli,CBMeMBer)滤波器.随后,在多机动目标时间演化模型和观测模型均为线性高斯的假设条件下利用高斯混合(Gaussianmixture,GM)技术获得了该滤波器解析的递推形式—多模型GM-CBMeMBer滤波器,并简要给出了它在非线性条件下的扩展卡尔曼(ExtendedKalman,EK)滤波近似.仿真实验结果表明所建议的多模型GM-CBMeMBer滤波器能有效地对多个机动目标进行跟踪而单模型GM-CBMeMBer滤波器则会产生明显的航迹丢失和虚假航迹,并且对于信噪比较低的仿真场景,它的性能优于多模型高斯混合概率假设密度(GMprobabilityhypothesisdensity,GM-PHD)滤波器,接近于多模型高斯混合势概率假设密度(GMcardinalizedPHD,GM-CPHD)滤波器.关键词引用格式DOI

多机动目标跟踪,势平衡多目标多伯努利滤波器,交互式多模型算法,高斯混合实现

连峰,韩崇昭,李晨.多模型GM-CBMeMBer滤波器及航迹形成.自动化学报,2014,40(2):336 34710.3724/SP.J.1004.2014.00336

Multiple-modelGM-CBMeMBerFilterandTrackContinuity

LIANFeng1

HANChong-Zhao1

LIChen2

AbstractAmulti-modelcardinalitybalancedmulti-targetmulti-Bernoulli(CBMeMBer) lterisproposedinthispaperfortrackingmultiplemaneuveringtargetsandformingthemulti-targettrajectoriesinclutter.Giventheassumptionsthatthedynamicandobservationmodelsofthemultimaneuveringtargetsarelinear-GaussianandbyapplyingtheGaussianmixture(GM)technique,theanalyticrecursionfortheproposed lter,namelythemulti-modelGM-CBMeMBer lter,isobtained.TheextendedKalman(EK) lteringapproximationsforthemulti-modelGM-CBMeMBer ltertoaccommodatenon-linearmodelsaredescribedbrie y.Simulationresultsshowthattheproposed lterperformsmultiplemaneuveringtargetstrackingwellwhereasthesingle-modelGM-CBMeMBer lterobviouslyproducesthemissingandfalsetrajectories.Inaddition,simulationresultsalsoshowthatforthescenariosoftherelativelylowsignal-to-noiseratio(SNR),theperformanceoftheproposed lterisbetterthanthatofthemulti-modelGMprobabilityhypothesisdensity(GM-PHD) lter,andisclosetothatofthemulti-modelGMcardinalizedPHD(GM-CPHD) lter.

KeywordsMultiplemaneuveringtargetstracking,cardinalitybalancedmulti-targetmulti-Bernoulli(CBMeMBer) lter,interactingmultiplemodels(IMM)algorithm,Gaussianmixture(GM)implementation

CitationLianFeng,HanChong-Zhao,LiChen.Multiple-modelGM-CBMeMBer lterandtrackcontinuity.ActaAutomaticaSinica,2014,40(2):336 347

收稿日期2012-07-20录用日期2013-01-28

ManuscriptreceivedJuly20,2012;acceptedJanuary28,2013国家重点基础研究发展计划(973计划)(2013CB329405),国家自然科学基金创新研究群体(61221063),中国博士后科学基金(20100481338),中国博士后科学基金特别资助项目(2012T50746),中央高校基本科研业务费专项资金资助

SupportedbyNationalBasicResearchProgramofChina(973Program)(2013CB329405),FoundationforInnovativeRe-searchGroupsoftheNationalNaturalScienceFoundationofChina(61221063),ChinaPostdoctoralScienceFoundation(20100481338),ChinaPostdoctoralScienceFoundation-SpecialFund(2012T50746),andFundamentalResearchFundsfortheCentralUniversity本文责任编委文成林

RecommendedbyAssociateEditorWENCheng-Lin

1.西安交通大学电信学院综合自动化研究所智能网络与网络安全教育部重点实验室西安7100492.西安交通大学软件学院西安7100491.MinistryofEducationKeyLaboratoryforIntelligentNet-worksandNetworkSecurity(MOEKLINNSLab),InstituteofIntegratedAutomation,SchoolofElectronicsandInforma-tionEngineering,Xi anJiaotongUniversity,Xi an7100492.SchoolofSoftwareEngineering,Xi anJiaotongUniversity,Xi an710049

近年来,基于随机有限集(Random niteset,RFS)[1]理论的多目标跟踪算法,尤其是概率假设密度(Probabilityhypothesisdensity,PHD)滤波器[2]和势概率假设密度(CardinalizedPHD,CPHD)滤波器[3 4],受到了国内外学者的广泛关注.Vo等分别给出了PHD和CPHD滤波器的序贯蒙特卡洛(SequentialMonteCarlo,SMC)实现[5],但该实现必须采用聚类算法从粒子群中提取多目标的状态估计,这一过程需要耗费昂贵的计算代价且所获得结果的可靠性较低[6 7].针对此问题,Vo等通过将多目标运动过程近似建模为多伯努利RFS,提出了势平衡多目标多伯努利(Cardinalitybal-ancedmulti-targetmulti-Bernoulli,CBMeMBer)滤波器[8]并给出了它的SMC实现以及在线性高斯条件下的高斯混合(Gaussianmixture,GM)实现.其中,相比SMC-PHD和SMC-CPHD滤波器,SMC-CBMeMBer滤波器的优势在于其不需要采用

第1页下一页