Person Detection Techniques for an Internet of Things(IoT...
Transcript of Person Detection Techniques for an Internet of Things(IoT...
PersonDetectionTechniquesforanInternetofThings(IoT)-BasedEmergencyEvacuationSystem
Prasad Annadata, PhD StudentWisam Eltarjaman, PhDRamkiThurimella,PhD,Professor
Contents
● Background● ProblemStatement● Motivation● Solution● Results● Conclusion
Settingthe Stage● FirekillsmoreAmericansthanallnaturaldisasterscombined.Everyyearmorethan5,000peopledieinfires,over25,000areinjuredanddirectpropertylossisestimatedatover$9billion[Federal Emergency Management Agency (FEMA)]
● OSHAstandardsexplicitlyrequiretheemployertotrainemployeesinsafetyandhealthaspectsoftheirjobs[Occupational Safety and Health Administration (OSHA), USA]
● FireProtectionSystemsMarketworth98.24BillionUSDby2022[Markets and Markets]
EmergencyPreparedness
● Emergencypreparednessistheanswer
● Ethicalresponsibilityofemployers
● Compliancerequirementinmanyjurisdictions
● Itcoststimeandresources
Notallemergenciescanbeavoidedorpredicted
TypicalSetup
● Initialinstallationandtraining
● Periodicmaintenanceofsafetyequipment
● Trainingofvolunteeremployeestobecomeemergencyresponsecoordinators(ERCs)
● Periodicpreparednessdrills
DuringanEmergency
Simulateevacuationascloselyaspossibletoarealemergency
ERCsperformcoordination• Emergencyresponders
ü Grabattention(whistles,signs)
ü Directeveryonetoassemblypoint(AP)
ü Ensureeveryonehasleftthebuilding
ü Accountforeveryoneattheassemblypoint
Contents
● Background● ProblemStatement● Motivation● Solution● Results● Conclusion
ProblemswithFireDrills
● ERCsandemployeeshaveamplenotice
● ERCspre-preparebyprintingattendancelists
● Reconciliationprocessistimeconsuming
● Createslackoffaithintheprocess
LifeisPriorityOne
● Savinglivesisthemostimportantpriorityduringanemergency
● Accountingforeverylifeiscrucial
● Attendancesystemsareusedtogetabaseline
● ManualreconciliationdoneattheAP
ProblemswithManualReconciliation
● Baselineaccuracyisdirectlydependentontheaccuracyofattendancesystems
● Attendancerequires− Registration− specific“arrivalaction”− Notsuitableforpublicareassuchasmalls− Consumesvaluableresponsetime− ParticularlyinlargeevacuationswithmultipleERCsandAPs
Contents
● Background● ProblemStatement● Motivation● Solution● Results● Conclusion
Impacts of Reconciliation Errors
FalseNegatives− Wronglycountingsomeoneas“Not-trapped”
− Mostdangeroustolives
FalsePositives− Wronglycountingalreadysafepeopleas“trapped”
− Wastesvaluableresourcesinsearching
AdvantagesofProposedIoT-BasedSolution
● ImproveAccuracyofAttendance− Automatic− Supplementsexistingattendancesystem
● AutomateReconciliation− AutomaticrecognitionofpersonnelatAPs− Ad-hocnetworktopreventduplicates− Quickproductionofstill-trappedreport
● Augmentreportwithlocation
Contents
● Background● ProblemStatement● Motivation● Solution● Results● Conclusion
What’stheIdea
● PeopleCarryMultipleDetectableDevices− Smartphonewithmultiplechannels− RFID-basedidentitycard− WearableswithBluetooth
● Eachhas“uniqueenough”ID● UseIoT-baseddevicesplacedaroundthebuildingtodetecttheseandcountthem
WhyIoT?
Enoughresearch
inIoT-basedad-hoc,robust
networkscrucialduringdisasters
MobileapplicationsforERCs
Low-costcommerciallyavailablehardware
Personaldevicescanbe
consideredIoT devices
Setup
● IoT sensors- placedacrossthebuilding(s)− Single-boardcomputer(Raspberry-Pi)
● Everysensorhasoneormorechannelsitcandetect,e.g.RFID,WiFi,Bluetooth
● Softwareisbuiltwith− Fault-tolerance− Integrationwithattendancesystems− IntegrationwithmobileappsforERCs
ExampleEarl70-yearold,securitydesk,carriesasingledumbphoneandacompanyIDcard
Betty45-yearold,adminrole,carriesasmartphoneandacompanyIDcard
Example
Ezra28-yearold,softwaredeveloper,carriesmultipledevices
Grace8-yearold,visitingherdad.CarriesnoIDatall
DetectionofMovement
● SameIDdetectedbydifferentIoT devices
● InitiallyeachmovingIDisassumedtobeadifferentperson
● Exit/Entrancenodesdospecialprocessing–subtract/addpersoncounts
MotionDetection- Camera
1.Motionisdetected
2.TrytodetecttheID
● IftheIDdetectedisknown− Noaction
MotionDetection- Camera
1.Movementisdetected
2.TrytodetecttheID
● NewIDdetected− Addperson
MotionDetection- Camera
1.MovementisDetected
2.TrytodetecttheID
● No ID detected− Electronically silent person
Co-occurringIDs
● NeedtodetectandmergeIDsbelongingtothesameperson
● Pairsofco-occurringIDsareenumerated● Iftheyco-occurmorethanthresholdnumberoftimes,thenmerge
● Brute-forceiscomputationallyintensive● Reverseindexingtechniqueusedtoimproveperformance
Co-Occurrence(StructuralEquivalence)
LetSi anddk denotesnapshotsanddeviceIDsrespectively.Fore.g.• S1 ={d2,d3,d7 },• S2 ={d3,d7,d8 },• S3 ={d2,d3,d7,d9 },...• S7={d1,d5,d6,d4,d2 }
S7
S3
S2
S1
d1
d2
d3
d4
d5
d7
d6
d9
d8
GroupingdeviceIDsthatco-occur
• n– numberofdevices,• m– numberofsnapshots• N– totalinputsize• Checkingifeverypairofdevicesco-occursisexpensive:(nC2)m• Ifsnapshotshavenoerror,i.e.alwaysdetectanID,optimalalgorithmispossible.
Co-OccurrenceAlgorithmAdjacencylistofleftpartition(invertedlist):d1={S7}d2 ={S1,S3,S7}d3 ={S1,S2,S3}d4={S7}andsoon
- d1 andd4 co-occur– canbedetectedefficientlyusingatrie datastructure- d2 andd3 approximately co-occur.Harderproblem.
- Jacard Index=- |N(d2)∩ N(d3)|/|N(d2)∪N(d3)|=2/6
S7
S3
S2
S1
d1
d2
d3
d4
d5
d7
d6
d9
d8
OtherTechniques
● Clean-upRoutines− Exitpersonsnotseenforawhile− Detectstaticitemsandremovethemfrompersonlist
− ExpirestaleIDs● Emergencytimeroutines− Exitnodescountexitedpersons− Reconciliationroutineskickedoff
Contents
● Background● ProblemStatement● Motivation● Solution● Results● Conclusion
ImplementationDetails
● Partoflargercomprehensivesolution● Simulationisused● Effortismadetomakeitrealistic− RealMACIDsused− Realhashingisused− Realisticdistributionsusedfor
● AssigningnumberofIDsperperson● EntranceandExitpatternsforpersons● IoT devicelocations
Results
● Timeintervalssimulated:36000● Numberofpeoplesimulated:1000● Numberoflocationsinthebuilding:1000
Results
● Parameterstunedtoeliminatefalsenegatives− Detectedcountisneverlowerthanactual
● Mostentrancesinthemorning&exitsintheevening(twobluelines– curveinthepreviousslideisnotsymmetricaroundtheverticalcenter)
Limitations
● Allpeoplehavesamespeedandneedofmovement
● Motion/Cameraarenotsimulated● Electronicallysilentpersondetectionisnotpossible
● AssumedperfectdetectionofIDsbysensorswheninrange(nonoiseintroduced)
● AssemblypointIDdetectionisnotdone● Reconciliationisnotsimulated
Contents● Background● ProblemStatement● Motivation● Solution● Results● Conclusion
Conclusion
● Presentedasetofsimpletechniquesthatenhancephysicalsecuritybysensingpersonsinbuildingsincludingtheirlocations
● Throughsimulationweshowedthatitisaviablepursuit
● Clearmathematicalmodelandalgorithmspresented(inthepaper)
● Savestime,moneyandmostimportantlylives
FutureDirection● ThesetechniquesbecomepartofacomprehensiveIoT-basedevacuationsolution
● Integrationandtestingwithrealattendancesystems
● Simulationofspecialsituations(e.g.movingassetssuchasprojectors)
● Extendthesystemtopublicspacesandpublicsafety
{ prasad, wisam, ramki }@cs.du.edu
Daniel Felix Ritchie School of Engineering & Computer Science
UNIVERSITY OF DENVER2155 East Wesley Avenue, Denver, CO
80208 - USA
http://crisp.cs.du.edu