Abstract—Visual-inertialSLAM(VI-SLAM)requiresagoodinitialestimationoftheinitialvelocity,orientationwithrespecttogravityandgyroscopeandaccelerometerbiases.InthispaperwebuildontheinitializationmethodproposedbyMartinelli[1]andextendedbyKaiseretal.[2],modifyingittobemoregeneralandefficient.Weimproveaccuracywithseveralroundsofvisual-inertialbundleadjustment,androbustifythemethodwithnovelobservabilityandconsensustests,thatdiscarderroneoussolutions.OurresultsontheEuRoCdatasetshowthat,whiletheoriginalmethodproducesscaleerrorsupto156%,ourmethodisabletoconsistentlyinitializeinlessthantwosecondswithscaleerrorsaround5%,whichcanbefurtherreducedtolessthan1%performingvisual-inertialbundleadjustmentaftertenseconds
2016/11/13 21:08:24
970KB
ICRA
1