Lecture 4 for cognitive interaction robo

in field Robotics

1. Respiration
1
•Bodycellsandtissuesneedoxygentolive.
•Respirationistheprocessthroughwhichtheoxygenneededfor
livingcellsisenteredintothelungsandthencirculated
throughoutthebody.
•Ithastwostages:
‒Inhalation
‒Exhalation

2. Respiration Cycle
2
Duringtheinhalation/exhalationcycle:
•oxygeniscarriedfromthelungsand
absorbedbytheredbloodcells(RBC).
•Hemoglobin(Hb)istheproteinthatcarries
oxygenintheRBCsandtransportsit
throughoutthebody.
•Theheartpumpsoxygenatedhemoglobin
(HbO
2)fromthelungstothewholebody
cellsandtissuesthroughthecircularity
system,andreceivesthedeoxygenated
Hbandpumpsittowardsthelungsagain
tobeoxygenated.

3. Cell Respiration Formula
3
•Thisformuladescribesthebiochemicalprocessin
whichthecellsofthebodyobtainenergyby
combiningoxygenandglucose,resultinginthe
releaseofcarbondioxide,water,andATP(Adenosine
Tri-Phosphate)
•ATPistheprimaryenergycarrierinlivingbeings.

4. Respiration Rate
4
•Therespirationrateisthenumberof
inhalation/exhalationcyclesinaminute,definedas
breathsperminuteandexpressedasBPM.
•Normalrespirationrateforadulthealthypersonsat
restrangesfrom12to20BPM,
•Foranewborn(<1year),itrangesfrom30to40
BPM.
•Thelowerorhigherrespirationrate,whileresting,is
consideredasbeingabnormal.

5. Respiration Rate Measurement
5
Respirationratemeasurementisachievedthrougheither
contactorcontactlessmethods.
1)Contactmeasurement
•Aphysicalelectronicsensorisattachedtothebody
skinortiedtothepatientclothestosensethe
motionofthechest.
•Theresultingmotionsignalisthenprocessedto
extracttherespirationrate.
•Thesemethodsarenotappropriateinsomecases
suchassensitiveorburnedskin,orforpremature
babies.
•Inaddition,contactsensorsarelessconvenient
andmoreexpensive.

6. Respiration Rate Measurement
6
2)Contactlessmeasurement
•Contactlessrespirationmethodsinvolvecomputervisiontechniques.
•Someexamplesincludethetrackingofcontractionandexpansionof
chestandabdomenregionwitheachinhalation/exhalationcycle.
•Also,Trackingoftemperaturevariationsaroundthenoseandmouth,
causedbyairflowmovementthatcouldbecapturedusingathermal
camera.
•Contactlessrespirationratemonitoringisfavorabletoovercomethe
issuesrelatedtoattachingsensorstosensitiveorburnedskin,or
tyinguncomfortablesensorstoclothes.
•Inaddition,contactlesstechniquesarepropersolutionsforsleeping-
relatedmonitoring.

7. RespirationSignal
7
Thefigurebelowshowsasampleofrespiratorysignalacquired
throughcomputervisionapproach.Thepeaksinthesignalmaybe
usedtoindicatebreathingcycles.

8. Application
8
•Implementacomputervision-basedapproachtoplotthe
breathingpatternforahumanduringhissleep.
Youcandownloadadatasetforthistaskfrom:
https://figshare.com/articles/dataset/sleep_dataset_zip/5518996/2

9. Face Recognition
9
•Faceisperhapsthemostcommonandfamiliarbiometricfeature.
•Thebroaduseofdigitalcamerasandsmartphonesmadefacial
imageseasytoproduceeveryday.
•Theseimagescanbeeasilydistributedandexchangedbyrapidly
establishedsocialnetworkssuchasFacebookandTwitter.
Face images

10. Face Recognition
10
•Thehumanfaceisnotanidealmodalitycomparedtoother
biometrictraits.
•Itislessprecisethanotherbiometricmodalitiessuchasirisor
fingerprint.
•Itcanpotentiallybeinfluencedbycosmetics,disguises,andlighting.
•However,thefacehastheadvantagesthatmakeitoneofthemost
favoredbiometriccharacteristicsforidentityrecognition.
-Naturalcharacter
-Nonintrusive
-Lesscooperation

11. Face Recognition History
11

12. Recent Advancement
12
•Initsnewupdates,Appleintroducedafacialrecognitionapplicationwhere
itsimplementationhasextendedtoretailandbanking.
•MastercarddevelopedtheSelfiePay,afacialrecognitionframeworkfor
onlinetransactions.
•From2019,peopleinChinawhowanttobuyanewphonewillnowconsent
tohavetheirfacescheckedbytheoperator.
•Chinesepoliceusedasmartmonitoringsystembasedonlivefacial
recognition;usingthissystem,theyarrestedasuspectof“economiccrime”
ataconcertwherehisface,listedinanationaldatabase,wasidentifiedina
crowdof50,000persons.
Today,facialrecognitiontechnologyadvancementhasencouragedmultiple
investmentsincommercial,industrial,andgovernmentalapplications.
Forexample:

13. Main Steps in Face Recognition Systems
13
Automatedfacerecognitionincludesthreekeysteps:
(1)Facedetection
(2)Extractionoffeatures
(3)Classification

14. Main Steps in Face Recognition Systems
14
1)Facedetection
Itisthefirststepintheautomatedfacerecognitionsystem.Itusually
determineswhetherornotanimageincludesaface(s).Ifitdoes,its
functionistotraceoneorseveralfacelocationsinthepicture.
2)Featureextraction
Thisstepconsistsofextractingfromthedetectedfaceafeaturevector
namedthesignature,whichmustbeenoughtorepresentaface.
3)Classification
•Canbeeitherverificationoridentification.
•Verificationrequiresmatchingonefacetoanothertoauthorizeaccessto
arequestedidentity.
•Identificationcomparesafacetoseveralotherfacesthataregivenwith
severalpossibilitiestofindtheface’sidentity.

15. Facial Landmarks
15
Afterdetectionofthefaceregioninthefirststep,whichresultsa
boundingboxthatsurroundthefaceintheimage,anumberof
featurepoints(landmarks)aredetermined(manuallyor
automatically)insidetheboundingbox.
Thereareusuallytwotypesoffaciallandmarks:
1)thefacialkeypoints:Thefacialkeypointsarethedominant
landmarksonface,suchastheeyecorners,nosetip,mouth
corners,etc.
2)interpolatedlandmarks:Theinterpolatedlandmarkpointseither
describethefacialcontourorconnectthekeypoints

16. Face Database
16
BioID:20landmarksareannotated
ibug:68landmarksareannotated
Heledominantn:194landmarksareannotated
MediaPipe:468landmarksareannotated

17. Face Recognition System
17
Duringaregistrationofanewface,anumberoflandmarksare
determinedautomaticallyormanually,andtheirlocationsarestored
forfuturematching.Theselandmarksarecalledgroundtruth
landmarks.
Duringvalidation,thesameprocedureisfollowedtoextractthe
landmarksandtheirlocations.
Amatchingscoreiscomputedbetweentheextractedlandmarksand
groundtruthlandmarks.
Thematchingscoredeterminesthedistancebetweeneach
landmarkanditscorrespondentgroundtruthlandmark.
Basedonthematchingscore,thetestedfaceimageisacceptedor
rejected.

18. Matching Score
18
•Thesimplestcomparisonisarootmeansquarederror(RMSE)assessment;
wheretheaveragedistancebetweeneachoftheNpredictedlandmarks
(&#3627408485;
??????
??????
,&#3627408486;
??????
??????
)andthecorresponding‘groundtruth’(&#3627408485;
??????
??????
,&#3627408486;
??????
??????
)iscalculatedonaper
landmarkbasis.
•Landmarksthatarepoorlypredictedwillbepositionedfaroftheir
correspondinggroundtruthlocationsandthuscontributetoincreasingthe
RMSEvalue.
&#3627408453;&#3627408448;&#3627408454;??????=
1
&#3627408449;
෍
??????=1
??????
&#3627408485;
??????
??????
−&#3627408485;
??????
??????
2
+&#3627408486;
??????
??????
−&#3627408486;
??????
??????
2

19. Applications
19
SolveApplication(1)anduploadittothecourseClassroom

20. 20