资源描述:
《Loosely Coupled Kalman Filtering for Fusion of》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、16thInternationalConferenceonInformationFusionIstanbul,Turkey,July9-12,2013LooselyCoupledKalmanFilteringforFusionofVisualOdometryandInertialNavigationSalimSırtkayaandBurakSeymenA.AydınAlatanASELSANInc.MiddleEastTechnicalUniversityMicroelectronicsGuidanceandElectro-opticsDivisionE
2、lectricalandElectronicsEngineeringDepartmentAnkara,TurkeyAnkara,TurkeyEmail:sirtkaya@aselsan.com.trEmail:alatan@eee.metu.edu.trbseymen@aselsan.com.trAbstractVisualOdometry(VO)istheprocessofestimat-inertialnavigation.Theresearchcommunityhasexploitedtheingthemotionofasystemusingsin
3、gleorstereocameras.benefitsofintegratingvisualandinertialsensorsinGPSdeniedPerformanceofVOiscomparabletothatofwheelodometersenvironments,andproposedsolutionsforseveralmissionssuchandGPSundercertainconditions;thereforeitisanacceptedasnavigation,pin-pointlandingandflightcontrol.choic
4、eforintegrationwithinertialnavigationsystemsespeciallyinGPSdeniedenvironments.Ingeneral,VOisintegratedwithThereareseveralwaystoutilizevisualmeasurementstheinertialsensorsinastateestimationframework.Despitethevariousinstancesofestimationfilters,theunderlyingconceptsformotionestimat
5、ion.StructurefromMotion(SfM),Si-remainthesame,anassumedkinematicmodelofthesystemismultaneousLocalizationandMapping(SLAM)andVisualcombinedwithmeasurementsofthestatesofthatsystem.TheOdometry(VO)areamongthepopularmethodologies.InSfMdrawbackofusingkinematicmodelsforstatetransitionist
6、hattheandSLAM,motionestimationisformulatedtogetherwithstateestimatewillonlybeasgoodastheprecisionofthemodel3Dstructureestimationofthecapturedenvironment.VO,inusedinthefilter.Acommonapproachinnavigationcommunityiscontrast,focusessolelyonmotionestimation.touseanerrorpropagationmodel
7、ofthenavigationsolutionusinginertialsensorinsteadofanassumeddynamicalmodel.HighrateForvisual-inertialintegration,visualdataisgenerallyin-IMUwilltracethedynamicbetterthananassumedmodel.Inthispaper,weproposealooselycoupledindirectfeedbackKalmancorporatedas2Dfeature(identifierpicture
8、element)matches.filterintegrationforvisua