双目立体视觉slam分析

《双目立体视觉slam分析》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、ABSTRACTABSTRACT：Therobotsimultaneouslocalizationandmapping(SLAM)isdeeplystudiedinthispaperwithsensorofBumblebee2thatmadeinCanadianPointGreycompany．Themainworkisasfollows：Firstly,thecameraimagemodel，binocularstereovision3．Dreconstruction，two-viewgeometry,aswellasfeaturedetection(Harris，

2、SURF)andmatchingalgorithmsareresearched．Bisides，thestereosystemiscalibratedbyMATLABtoolbox．Furthermore，theexperimentsofHarrisandSURFfeatureextractionandmatchingarestudiedinconditionofknownandunknownpolargeometricconstraint．Fromtheresults，theperformanceofHarrisandSURFareconcluded．Secondl

3、y,formobilerobotlocalizationinknownenvironment，the5％orderConjugateUnscentedParticleFilterMonteCarloLocalization(CUPF-MCL)algorithmisproposedbasedonthetheoryof5％orderConjugateUnscentedTransform(5mCUT)andParticleFilterMonteCarloLocalization(PF．MCL)．CUPF．MCLcombines5mCUTwithKalmanFilter(KF

4、)togeneratemoreaccuracyparticlefilterproposaldistribution，whichcalculatestheMCLtransitiondensityuptothe5％ordernonlinearity．Insimulation，CUPF—MCLiscomparedwithdeadreckoning，PF—MCL，EPF-MCLandUPF-MCL．TheresultsshowthatCUPF-MCLalgorithmovercomestheproblemofparticledegradationandimprovesthea

5、ccuracyofMCL．Thirdly,formobilerobotSLAMinunknownenvironment，thispaperproposesCUFastSLAMbasedon5mCUTandRao．BlackwellizedParticleFilter(RBPF)．Themaincharacteristicsofthealgorithmare"(1)aJacobian—freeconjugateunscentedparticlefilterWasderivedtoestimatetheposteriorsoftherobotstate；(2)revisi

6、tedenvironmentlandmarksareaccuratelyupdatedwithasetof5％orderconjugateunscentedKalmanfilters；(3)newlyvisitedenvironmentlandmarksareregisteredinthemapuptothe5th-orderofnonlinearitywithoutevaluatingthemeasurementJacobians．WedemonstratetheperformanceofCUFastSLAMwiththatofFastSLAM2．0andUFast

7、SLAMinsimulationsandexperiments．TheresultsshowthatCUFastSLAMoutperformsFastSLAM2．0andUFastSLAMbothinrobotlocalizationandenvironmentmapping．Finally,theSLAMexperimentbasedonbinocularstereovisioniscarriedoutinVll北京交通大学硕士学位论文CUFastSLAMframework，whichprovidesavaluablereferenceforthe