Ferret RFID Localization for Pervasive Multimedia

Ferret:RFIDLocalizationforPervasive

Multimedia󰀁

XiaotaoLiu,MarkD.Corner,andPrashantShenoy

DepartmentofComputerScienceUniversityofMassachusetts,Amherst,MA{xiaotaol,mcorner,shenoy}@cs.umass.edu

Abstract.Thepervasivenatureofmultimediarecordingdevicesen-ablesnovelpervasivemultimediaapplicationswithautomatic,inexpen-sive,andubiquitousidentificationandlocationingabilities.WepresentthedesignandimplementationofFerret,ascalablesystemforlocatingnomadicobjectsaugmentedwithRFIDtagsanddisplayingthemtoauserinreal-time.Wepresenttwoalternativealgorithmsforrefiningapostulationofanobject’slocationusingastreamofnoisyreadingsfromanRFIDreader:anonlinealgorithmforreal-timeuseonamobiledevice,andanofflinealgorithmforuseinpost-processingapplications.Wealsopresentmethodsfordetectingwhennomadicobjectsmoveandhowtoresetthealgorithmstorestarttherefinementprocess.AnexperimentalevaluationoftheFerretprototypeshowsthat(i)Ferretcanrefineobjectlocationstoonly1%ofthereader’scoverageregioninlessthan2min-uteswithsmallerrorrate(2.22%);(ii)Ferretcandetectnomadicobjectswith100%accuracywhenthenomadicdistancesexceed20cm;and(iii)Ferretworkswithavarietyofusermobilitypatterns.

1Introduction

Advancesindigitalimagingtechnologieshaveledtoaproliferationofconsumer

deviceswithvideocapturecapabilities.Thepervasivenatureofmultimedia

recordingdevicessuchascellphones,digitalcamcorders,PDAsandlaptops,

hasmadeitrelativelysimpletocapture,transform,andsharelargevolumes

ofpersonalvideoandimagecontent.Aconcurrenttrendistheemergenceof

low-costidentificationtechnologiessuchasRFIDtags,designedtoreplacebar-

codes[12].Eachtagcontainsanumericcodethatuniquelyidentifiestheobject

andcanbequeriedbyawirelessreader.Itislikelythatinthenearfuturemany

personalobjects(e.g.,books,clothing,fooditems,furniture)willbeequipped

withself-identifyingRFIDtags.

Theconfluenceofthesetrends—theubiquityofRFIDtagsandtheperva-

sivenatureofmultimediarecordingdevices—enablesnovelpervasivemultimedia

applicationswithautomatic,inexpensive,andubiquitousidentificationandlo-

cationabilities.ByequippingcameraswithRFIDreaders,itispossibletorecord

󰀁ThisresearchwassupportedinpartbyNSFgrantsNationalScienceFoundationgrantsCNS-0219520,CNS-052072andEIA-0098060.imagesaswellastheidentitiesandlocationsofallRFID-taggedobjectscon-

tainedwithineachimage.Thecapturedvideocanthenbequeriedinreal-time

todisplaythelocationofaparticularobject.

WhiletheinexpensivenatureofRFIDtagseaseslarge-scaledeploymentis-

sues,theirpassivenatureraisesanumberofhurdles.Akeylimitationisthatpas-

siveRFIDtagsareself-identifyingbutnotself-locating(i.e.,uponbeingqueried,

atagcanreportitsidentifybutnotitslocation).Consequently,ifmultipleob-

jectsarepresentinacapturedimage,itisnotpossibletodistinguishbetween

theseobjectsorpinpointtheirindividuallocations.Someoftheapplications

(e.g.,pinpointingamisplacedbookonabookshelf)requirelocationinformation

inadditiontoobjectidentities.Whilenumerouslocationingtechnologiessuch

asGPSandultrasound[10,13,14]areavailable,itisnotpossibletoequippas-

siveRFIDtagswiththesecapabilitiesduetoreasonsofcost,form-factorand

limitedbatterylife.Instead,werequirealocationingtechnologythatdoesnot

dependonmodificationstotags,iseasilymaintained,andscalestohundredsor

thousandsoftaggedobjects.

Toaddresstheabovechallenges,wehavedesignedasystemcalledFerret.

Ferretcombineslocationingtechnologieswithpervasivemultimediaapplications.

FerretcanlocateobjectsusingtheirRFIDtagsanddisplaystheirlocationsin

real-timetoamobileuser.Asthepositionsofobjectsareuncertain,thesystem

overlaysthevideodisplaywithanoutlineofwheretheobjectprobablyis.For

instance,auserwithaportablecameracanaskthesystemtodisplaythelocation

ofeverynewobjectintheroom,andthedisplaywillshowanoutlineofallof

thoselocations.Thedisplayisconstantlyupdatedastheusermovesusingthe

continuousstreamoftagreadingstoupdatelocations.

FerretusesthelocationanddirectionalityofRFIDreaderstoinferthelo-

cationsofnearbytags.Ferretleveragestheuser’sinherentmobilitytoproduce

readingsofthetagfrommultiplevantagepoints.Itdoesthisthroughtwonovel

algorithmsthatrefinethelocationsofobjectsusingastreamofnoisyreadings

fromRFIDtags.Onealgorithmisdesignedforofflineuse,givenalargeamount

ofcomputationalpower,whiletheotherisdesignedtooperateinreal-timeon

amobilesystem.Inthecaseoftheofflinealgorithm,wealsoincorporateneg-

ativereadings—whenthereaderdoesnotseetheobject—thisgreatlyreduces

theobject’spossiblelocations.

WehaveimplementedaprototypeofFerretandhaveusedittoconducta

detailedperformanceevaluation.Ourexperimentspayspecificattentiontohow

fastFerretcanrefineobjectlocations,theerrorrateinlocatingobjects,and

howwellithandlesnomadicobjects.Ourresultsshowthat(i)Ferretcanrefine

objectlocationstoonly1%ofthereader’scoverageregioninlessthan2minutes

withsmallerrorrate(2.22%);(ii)Theofflinealgorithmincorporatesinformation

aboutnotseeingtheobject,outperformingtheonlinealgorithmbyafactorof

13ormore;(iii)Ferretcandetectnomadicobjectswith100%accuracywhen

themovingdistancesexceed20cm;and(iv)Ferretworkswithawidevarietyof

usermobilitypatterns.

Ferretprovidessystemssupportforavarietyofnewapplications.Forin-

stance,userscanlocateamisplacedbookonabookshelf.Robotscanusesuch