通过融合视觉惯性特征实现身份感知跟踪.pdfVIP

2019InternationalConferenceonRoboticsandAutomation(ICRA)

PalaisdescongresdeMontreal,Montreal,Cay20-24,2019

EnablingIdentity-AwareTrackingviaFusionofVisualandInertialFeatures

RichardYi-ChiaTsai,HansTing-YuanKe,KateChing-JuLin,andYu-CheeTseng

DepartmentofComputerScience,NationalChiaoTungUniversity,Hsinchu,,R.O.C.

{yctsai5865,prelude0390,katelin,yctseng}@cs.nctu.edu.tw

789()#%*#2%:%/;$/234=50?)*0

—identificationandtracking(PIT)isan

essentialissueincomputervisionandroboticapplications.

Ithaslongbeenstudiedandachievedbytechnologiessuch

asRFIDorface/fingerprint/irisrecognition.Theseapproaches,

however,havetheirlimitationsduetoenvironmentalconstraints

(suchaslightingandobstacles)orrequireclosecontactto!!#D!$

specificdevices.Therefore,theirrecognitionaccuracyhighly,(-).)/0.0123450/06

dependsonusescenarios.Inthiswork,weproposeRCU

(RobotCatchyoU),anaccompanyistrobotsystemthatprovides

follow-meorguide-meservices.Suchrobotsarecapableof

distinguishingusers’profilesinfrontofthemandkeeptracking#D(

aspecifictarget.Westudyamorechallengingscenario,7)*/$0.50/06!#$%()*+)*

wherethetargetmaybeunderocclusionfromtimeto

time.ToenablerobustPIT,wedevelopadatafusiontechniqueFig.1:ThesystemarchitectureofRCU.

thatintegratestwotypesofsensors,anRGB-Dcameraand

wearableinertialsensors.Sincethedatag