资源描述:
《A Sequential Monte Carlo Method for Particle Filters》由会员上传分享,免费在线阅读,更多相关内容在行业资料-天天文库。
1、ASequentialMonteCarloMethodforParticleFiltersHongzhiGaoandRichardGreenDepartmentofComputerScienceandSoftwareEngineering,UniversityofCanterbury,ChristchurchNewZealand.{honghzi.gao,richard.green}@canterbury.ac.nzAbstractAnobjectorientedparticlefilterframeworkisproposedba
2、sedonsequentialMonteCarlomethods.Particlefilterisanextensivelyusedalgorithmforvisionbasedtrackingsystems.However,littleworkhasbeendoneinthepastliteraturetoinvestigatetheimplementationstrategiesoftheparticlefilteralgorithm.Inthispaper,weproposeaframeworkbasedonopensourc
3、eparticlefilterlibrariesandevaluaterespectiveadvantagesanddisadvantages.Theresultssupporttheproposedobjectorientedparticlefilterbeingamostusefultoolforcomputervisionbasedstochasticprediction.Keywords:particlefilter,implementationstrategy,applicationframework1.Introduct
4、ionThispaperwillinvestigatetheimplementationstrategiesoftheparticlefilterandquantitativelyParticlefilterisanonparametricalternativetoevaluatehowtheseimplementationissuesaffectitsGaussianbasedtechniques,suchasKalmanfilter[1],trackingaccuracy.whichprovidesatractableimple
5、mentationoftheBayesfilter[2]indecomposedstatespace.Insteadofrelyingonafixedfunctionalformoftheposterior,theInthenextsection,wediscusssomeopensourcekeyideaoftheparticlefilteristoapproximatetheparticlefilterimplementations.Insectionthree,theposteriorbeliefܾ݈݁ሺݔ௧ሻbyafinit
6、enumberofsamples,overallstructureoftheparticlefilterisintroduced,whicharerandomlydrawnfromthisposteriorfollowedbythedetailsofourimplementation.Inሾଵሿሾଶሿሾெሿsectionfour,weevaluateourimplementationofthedistributionanddenotedasΧ௧ൌݔ௧,ݔ௧,……,ݔ௧particlefilterquantitativelyandd
7、iscusshowthe[3].accuracyandefficiencyoftheparticlefilterareaffectedbyimplementationdecisions,suchaschoosingofparameters.WewillconcludethispaperComparingwithparametricbasedBayesfilterinsectionfive.implementations,particlefilterhasfollowingadvantages[3]:Firstly,particlef
8、ilterhasan‘any-time’[3]characteristic,whichenablesdesignerstotradeoffaccuracywithcomputationalefficiency.Secondly,2.R