Modeling and assessing quality of information in m

ModelingandAssessingQualityofInformationinMulti-sensorMultimediaMonitoringSystemsM.AnwarHossain1,2,PradeepK.Atrey3andAbdulmotalebElSaddik11MultimediaCommunicationsResearchLaboratory,UniversityofOttawa,Canada2CCIS,SoftwareEngineeringDepartment,KingSaudUniversity,Riyadh,KSA3DepartmentofAppliedComputerScience,TheUniversityofWinnipeg,CanadaCurrentsensor-basedmonitoringsystemsusemultiplesensorsinordertoidentifyhigh-levelinformationbasedontheeventsthattakeplaceinthemonitoredenvironment.Thisinformationisobtainedthroughlowlevelprocessingofsensorymediastreams,whichareusuallynoisyandimprecise,leadingtomanyundesiredconsequencessuchasfalsealarms,serviceinterruptions,andoftenviolationofprivacy.Therefore,weneedamechanismtocomputethequalityofsensor-driveninformationthatwouldhelpauserorasysteminmakinganinformeddecisionandimprovetheautomatedmonitoringprocess.Inthispaper,weproposeamodeltocharacterizesuchqualityofinformationinamulti-sensormultimediamonitoringsystemintermsofcertainty,accuracy/confidenceandtimeliness.Ourmodeladoptsamulti-modalfusionapproachtoobtainthetargetinformationanddynamicallycomputetheseattributesbasedontheobservationsoftheparticipatingsensors.Weconsidertheenvironmentcontext,theagreement/disagreementamongthesensors,andtheirpriorconfidenceinthefusionprocessindeterminingtheinformationofinterest.Theproposedmethodisdemonstratedbydevelopinganddeployingareal-timemonitoringsysteminasimulatedsmartenvironment.Theeffectivenessandsuitabilityofthemethodhasbeendemonstratedbydynamicallyassessingthevalueofthethreequalityattributeswithrespecttothedetectionandidentificationofhumanpresenceintheenvironment.CategoriesandSubjectDescriptors:...[...]:...GeneralTerms:...AdditionalKeyWordsandPhrases:Qualityofinformation,multimediamonitoringsystems,eventdetection,multi-modalfusion1.INTRODUCTIONMulti-sensormultimediasystemsarebecomingincreasinglypopularandarebeingdeployedinhomes,buses,shop-pingmalls,schools,andotherpremisesforvarioussurveillanceandmonitoringpurposes.Thesesystemsusuallyrecord,process,andanalyzethesensorymediastreamsinareal-timeoroff-linefashioninordertoprovidehigh-levelinformationaboutthehappeningsintheenvironment.Forexample,inanambientassistedlivingspace,multiplesensors(e.g.audio,video,RFID,proximity)aredeployedtoidentifythecurrentactivityofanelderlytoensurehissafetyandprovidehimwithtimelyassistance[Chenetal.2007].Therefore,theeffectivenessofthesesystemshighlydependsonthequalityofinformationtheyprovide.Thequalityofinformation(QoI)maybeexpressedasthemeasureofgoodnessofthehigh-levelinformation,whichwouldallowanautomatedsystemoradecisionmakertotakeaninformeddecisionabouttheeventsoccurringinthemonitoredenvironment.Typicallyamulti-sensorsystemobtainshigh-levelinformationofinterestbyfusingtheelementarydataprovidedbytheindividualsensors[Wald1999].Theuseofmultiplesensorsandthedataofdifferentmodalitiesoftenprovidesbetterresultsinobtainingtheinformationofinterestthanusingasinglesensor[Atreyetal.2006;Snidaroetal.2007].Theinformationobtainedinthismanneristhenusedforvarioussafetyandmonitoringpurposes.However,sensormeasurementsandprocessingareusuallynoisyandimprecise[Nakamuraetal.2007;Carvalhoetal.2003;PengandCandan2005].ThisputsemphasisonthecomputationofQoI,asthelowqualityofsuchinformationmayleadtofalsealarms,harassments,violationofprivacy,interruptioninservices,orotherundesiredconsequences.Inpreviousapproaches,severalattributesincludingtimeliness,accuracy,confidence,throughputandcosthavebeen...Permissiontomakedigital/hardcopyofallorpartofthismaterialwithoutfeeforpersonalorclassroomuseprovidedthatthecopiesarenotmadeordistributedforprofitorcommercialadvantage,theACMcopyright/servernotice,thetitleofthepublication,anditsdateappear,andnoticeisgiventhatcopyingisbypermissionoftheACM,Inc.Tocopyotherwise,torepublish,topostonservers,ortoredistributetolistsrequirespriorspecificpermissionand/orafee.c202011ACM1529-3785/202011/0700-0001$5.00ACMTransactionsonMultimediaComputing,CommunicationsandApplications,Vol.7,No.1,2011,2·...proposedspecificallyforestimatingthequalityofsensorydata[BlaschandPlano2005].Someoftheseattributeshavealsobeenusedforassessingthequalityofhigh-levelinformationbasedontheprocessingoflowlevelsensorydata[Bisdikian2007;Zilianietal.2005],howeverinalimitedscopeandfocus.Inthedomainoftraditionalinformationsystems,attributessuchasaccuracy,timeliness,precision,completeness,usabilityandmanyotherfactorshavebeenadvocatedforinformationqualitymeasure[Ehikioya1999;WangandStrong1996;Miller1996].Whiletheseap-proachesindicatetheuseofseveralqualityattributes,theyhavenotpresentedanymechanismonhowtocomputethedifferentqualityattributesbasedonmultiplesensorymedia.Incontrast,wedefineathree-attributeQoImetricinthispaperandthesearecertainty,accuracy/confidenceandtimeliness.Here,wehaveignoredthroughputfromtheQoIattributeduetoitsbeingmorerelevanttosensorydatadeliveryratethantohigh-levelinformation.ThecostattributeisnotofprimaryimportancewhencomputingQoIratheritisrelatedtodatacollection,processing,anddelivery.Theprecisionattributeindicatesthelevelofdetailin