Introduction to Eye Tracking and Tobii Eye Trackers

8/14/2019 Introduction to Eye Tracking and Tobii Eye Trackers

1/14

Tobii Technology

Tobii Eye TrackingAn introduction to eye tracking

and Tobii Eye Trackers

WhitePaper


2/14


3/14

TobiiEyeTrackingAn introduction to eye tracking

nd Tobii Eye Trackersa January 27, 2010

Tobii Technology AB

Eye tracking commonly refers to the technique used to record and measure

eye movements. In the last two to three decades we have witnessed a rapid

evolution in eye tracking technology with systems becoming easier to

operate and less intrusive to the test subjects. As a consequence the user

base has also expanded with eye tracking being used more commonly in

different research and commercial projects. The aim of this paper is to give

a brief introduction to the human visual system, and to explain how eye

movements are recorded and processed by Tobii Eye Trackers. Some basic

concepts and issues related to remote eye tracking and eye movement data

interpretation are also briefly discussed.

Fileversion:Tobii_Whitepaper_TobiiEyeTrackingWhitepaper_270110.pdf


4/14

2010TobiiTechnology Allrightsreserved.TobiiTechnologyandtheTobiilogoareeitherregisteredtrademarksortrademarksofTobii

TechnologyintheUnitedStatesand/orothercountries.

www.tobii.com

2

C

1 Whystudyeyemovements?....................................................................................................................3

ontents

1.1 Howdoestheeyework?...................................................................................................................3

1.2 Whydooureyesmove?....................................................................................................................4

1.3 Whatisvisualattention?...................................................................................................................5

1.4 Howfastishumanvisualperception?...............................................................................................5

1.5 Whatdowestudywhenweuseeyetrackingdata?.........................................................................6

2 HowdoTobiiEyeTrackerswork?............................................................................................................6

2.1 Howareeyemovementstracked?....................................................................................................6

2.2 WhatareDarkandBrightPupileyetracking?...................................................................................7

2.3 Whathappensduringthecalibration?..............................................................................................7

2.4 Howarefixationsdefinedwhenanalyzingeyetrackingdata?.........................................................8

2.5 IspupilsizecalculationpossiblewithTobiiEyeTrackers?................................................................8

2.6 Howdoesblinkingaffecteyetracking?.............................................................................................9

2.7 Doesheadmovementaffect eyetrackingresults?..........................................................................9

3 WhatinfluencestheaccuracyofaTobiiEyeTracker?.............................................................................9

3.1 Eyemovements.................................................................................................................................9

3.2 Drift..................................................................................................................................................10

4

Conclusion..............................................................................................................................................11

Whatdoeseyetrackingdatatellus?.....................................................................................................11

5

6 Bibliographicsources.............................................................................................................................12


5/14

2010TobiiTechnology Allrightsreserved.TobiiTechnologyandtheTobiilogoareeitherregisteredtrademarksortrademarksof

TobiiTechnologyintheUnitedStatesand/orothercountries.

www.tobii.com

3

1 Whystudyeyemovements?In order to understand the reasoning behind

studyingeyemovements, somebasic factsabout

thehumanvisionneed tobeknown.Thissection

provides short explanations of important terms

andcharacteristics

of

human

vision.

1.1 Howdoestheeyework?Oureyeshavemanysimilaritieswithhowaphoto

cameraworks:Lightreflectedfromanobjectora

scene travels into our eyes through a lens. This

lens concentrates and projects the light on to a

light sensitive surface located on the back of a

closed chamber. However, unlike a camera, the

light sensitive surface (which in the eye is called

the

retina,

see

Figure

2)

is

not

equally

sensitive

everywhere. Through evolution, our eyes have

been designed to work in both dark and light

environmentsaswellasprovidingbothdetailand

quick changes in what we see. This has led to

certain compromises, e.g., that we can only see

detailsclearlyinalimitedpartofourvisualfield(in

theeyecalled thefovealarea).The largerpartof

our visual field (the peripheral area) is better

adapted to low light vision, and to detect

movementsandregistercontrastsbetweencolors

andshapes.

The

image

produced

by

this

area

is

blurryandlesscolorful. Betweenthesetwoareas

wefindaregionoftransitioncalledtheparafoveal

area, inwhichthe imagebecomesgraduallymore

blurry as we move from the fovea into the

peripheralarea(seeFigure1).

The causeof thedifferences inour visual field is

the two different kinds of light receptor cells

available in the eye, i.e. the rods and the cone

cells.About 94%of the receptor cells in the eye

arerods.

As

mentioned

previously,

the

peripheral

areaof the retina isnot very good at registering

color and providing a sharp image of theworld.

Thisisbecausethisareaismostlycoveredbyrods.

Rodsdonot requiremuch light inorder towork,

butdo,ontheotherhand,onlyprovideablurred

andcolorlessimageofoursurroundings.Formore

detailed and clear vision, our eyes are also

equipped with light receptor cells called cones

whichmakeupabout6%of the totalnumberof

light receptor cells inoureyes.Conesare, in the

humaneye,

most

often

available

in

three

different

varieties;one that registersblue colors,one that

registersgreen andone that registers red.While

being efficient in providing a clear picture, the

cones do require much more light in order to

function.Hence,whenwelookatthingswhenits

darkaroundus,welosetheabilitytoseecolorand

use mainly information registered by rods,

providing uswith a grey scale image. Cones are

mostly found within the fovea where they are

tightly

packed

in

order

to

provide

as

clear

an

imageaspossible.

Figure2 ThehumaneyeTheretinaisalightsensitivestructureinsideoftheeye

responsiblefortransforming lightintosignals,whicharelater

convertedintoanimagebythevisualcortexinthebrain. The

oveaisasectionoftheretinathatcontainsahighdensityofboth

kindsoflightreceptorcellsfoundintheeye,i.e.ConeandRod

cells.Rodcells,whicharemostlylocatedintheouterretina,have

lowspatial

resolution,

support

vision

in

low

light

conditions,

do

notdiscriminate colors,aresensitivetoobjectmovementandare

responsiblefortheperipheralvision.Conecells,whicharedensely

packedwithinthecentralvisualfield,functionbestinbrightlight,

processacuteimagesanddiscriminate colors.

Figure1ThehumanvisualfieldThisfigureisaschematicrepresentationofthehumanvisual

ield.Themainareathatisinfocus,F,correspondstothearea

wherewedirectourgazetothefovealarea.Asisillustratedin

thisimage,thefovealareaisnotcircular.Hence,theareain

ocuswillhaveaslightlyirregularshapeaswell.Withintherest

ofthe

visual

field

(the

para

foveal

and

peripheral

areas)

the

imageweperceiveisblurryandthushardertointerpretand

discriminate inhighdetail.


6/14



www.tobii.com

4

1.2 Whydooureyesmove?The human visual field spans about 220 degrees

andis,aspreviouslymentioned,dividedin3main

regions:foveal,parafoveal,andperipheralregion.

We primarily register visual data through the

fovealregion

(Figure

1and

Figure

2)

which

constitutes less than 8% of the visual field. Even

though this represents only a small part of our

fieldofvision, the informationregisteredthrough

the foveal region constitutes50%ofwhat is sent

to the brain through our optic nerve. Our

peripheralvisionhasa verypooracuity,which is

illustratedinFigure1,andisonlygoodforpicking

upmovementsandcontrasts.Thuswhenwemove

oureyestofocusonaspecificregionofan image

or

object,

we

are

essentially

placing

the

foveal

region of the eye on top of the area which is

currentlywithinmainfocusofthelens inoureye.

Thismeans thatweare consequentlymaximizing

ourvisualprocessing resourceson thatparticular

area of the visual field which also has the best

imagedue to theopticcharacteristicsof theeye.

Bylettingthefovealregionregistertheimage,the

brain get the highest resolution possible for the

imageoftheinterestingareatoprocessaswellas

the most amount of data registered by the eye

aboutthat

area.

Hence,

the

brain

is

able

to

present the best possible image of the area we

findinterestingtous.

Eyemovementshave3main functionswhichare

considered important when we process visual

information:

1. Placetheinformationthatinterestsusonthe fovea. To do this, fixations and

saccadesareused.Afixationisthepause

ofthe

eye

movement

on

aspecific

area

of

the visual field; and saccades the rapid

movementsbetweenfixations.

2. Keeptheimagestationaryontheretinainspiteofmovementsoftheobjectorones

head.Thismovement iscommonlycalled

asmoothpursuit.

3. Prevent stationary objects from fadingperceptually. Movements used for this

are called microsaccades, tremors and

drift.

WhatareVisualAngles?

Oftenwhenwe readarticlesonhuman visionand

eye

tracking

accuracy,

we

come

across

measurements expressed in visual angle, e.g. the

sizeofthefovealareaisestimatedtobe12visual

angle, or remote eye trackers have an accuracy

between10.5. (Note:A smalleranglemeans less

inaccuracy.)Whenwepointaflashlightonawallin

adarkroomwecanobserve that the light formsa

projection on thewall. The size and shape of this

projection is related to the sizeof the light source

and thedistancethatyoustandfromthewall.The

reason

the

distance

affects

the

size

of

the

projection

is because the light disperses at a specific angle

from the source.Hence, ifwewish to specify the

size of the projection area using a standard size

measure(e.g.cmorcm2)ofthatflashlightwewould

alwayshavetospecifythedistanceatwhich itwas

measured. However, if we use the angle of

dispersionas size indicatorwe caneasily calculate

the projection size for multiple distances using

simpletrigonometry.Thesamerationalityappliesto

our visual field as images are formed through the

projectionof

light

on

the

retina,

i.e.

our

eye

works

asareversedflashlightthatabsorbslightinsteadof

emittingit.


7/14



www.tobii.com

5

1.3 Whatisvisualattention?Wheneverwelookattheworld,weconsciouslyor

unconsciouslyfocus

only

on

afraction

of

the

total

information thatwe couldpotentially process, in

other words we perform a perceptual selection

process called attention. Visually this is most

commonly done by moving our eyes from one

placeofthevisualfieldtoanother;thisprocess is

oftenreferredtoasachange inovertattention

ourgaze followsourattention shift.Even though

weprefertomoveoureyestoshiftourattention,

wearealsocapabletomoveourmindsattention

to theperipheralareasofourvisual fieldwithout

eyemovements (seeFigure1).Thismechanism is

calledcovertattention.Althoughwecanusethese

twomechanisms separately theymost frequently

occur together. An example is when we are

looking at a city landscape andwe first use our

covertattentiontodetectashapeormovementin

ourvisual fieldthatappearstobe interestingand

useourperipheralvision toroughly identifywhat

it is. We then direct our gaze to that location

allowing our brain to access more detailed

information.Thus

ashift

of

our

overall

attention

is

commonlyinitiatedbyourcovertattentionquickly

followedbyashiftofourovertattentionandthe

correspondingeyemovements.

1.4 Howfastishumanvisualperception?

Inadditiontoonlyhavingaverylimitedsharpfield

ofvision,oureyesarealsofairlyslowatregistering

changes in images compared to the update

frequencyofamoderncomputerscreen.Research

has shown that the retinaneeds about80msof

seeinganewimagebeforethatimageisregistered

innormal lightconditions.Thisdoesntmeanthat

we consciously have noticed any changes only

thattheeyehasregisteredachange.Theabilityto

register an image is also dependent on the light

intensityofthatimage.Thiscanbecomparedwith

a photographic camera where a short shutter

speed, inabadly litenvironmentresults inadark

andblurred image,wherehardlyanythingcanbe

seen. However, if taking an image of something

which is verywell lit, e.g. awindow, the shutter

speed can be very short without this problem

occurring. Inaddition toneeding time to register

animage,

the

eye

also

requires

time

for

the

image

to disappear from the retina. This is also

dependenton the light intensity.Oneexampleof

Figure3SceneperceptionRatherthanperceivinganobjectorasceneasawholewefixateonrelevantfeaturesthatattractourvisualattention,

andconstructthesceneinourvisualcortexusingtheinformationacquiredduringthosefixations.


8/14



www.tobii.com

6

this iswhenbeing exposed to a very bright light

such as a camera flash where the image of the

flashstaysontheretinalongaftertheflashinghas

ended.

In

addition

to

the

light

sensitivity

of

the

eye,

how

fastweperceivesomethingwearelookingatalso

dependsonwhatweareobserving.Whenreading

in normal light conditions, it has been observed

thatmostpeopleonlyneedbetween5060msof

seeing a word in order to perceive it. However,

whenlookingat,e.g.,apicturepeopleneedtosee

it for more than 150 ms before being able to

interpretwhattheyareseeing.

1.5 Whatdowestudywhenweuseeyetrackingdata?

Most eye tracking studies aim to identify and

analyzepatternsofvisualattentionof individuals

when performing specific tasks (e.g. reading,

searching, scanning an image, driving, etc.). In

these studies eye movements are typically

analyzedintermsoffixationsandsaccades.During

eachsaccadevisualacuity issuppressedand,asa

result,weareunabletoseeatall.Weperceivethe

world visually only through fixations. The brain

virtually

integrates

the

visual

images

that

we

acquire through successive fixations into a visual

sceneorobject(seeFigure3).Furthermoreweare

only able to combine features into an accurate

perceptionwhenwefixateandfocusourattention

on them. The more complicated, confusing or

interestingthosefeaturesarethe longerweneed

toprocess themand, consequently,more time is

spentfixatingonthem.Inmostcaseswecanonly

perceiveandinterpretsomethingclearlywhenwe

fixateonanobjectorareveryclosetoit.Thiseye

mindrelationship

is

what

makes

it

possible

to

use

eye movement measurements to tell something

abouthumanbehavior.

2 HowdoTobiiEyeTrackerswork?

The process of eye tracking is, from a technical

point of view, divided into two different parts:

registering the eye movements and presenting

them to theuser inameaningfulway.While the

eyetracker

records

the

eye

movements

sample

by

sample, the software runningon thecomputer is

responsible for interpreting the fixations within

the data. This chapter is about what happens

during an eye tracking process from a technical

point of view and aims to answer a few of the

questionsthatariseregardingthis.

2.1 Howaretheeyemovementstracked?

Eye trackinghas longbeenknownandusedasa

methodtostudythevisualattentionofindividuals.

There are several different techniques to detect

and track themovements of the eyes.However,

when it comes to remote, nonintrusive, eye

tracking the most commonly used technique is

PupilCentreCornealReflection (PCCR).Thebasic

concept is tousea light source to illuminate the

eyecausinghighlyvisiblereflections,andacamera

to capture an image of the eye showing these

reflections.The image capturedby the camera is

then used to identify the reflection of the light

sourceon thecornea (glint)and in thepupil (See

figure4).

We

are

then

able

to

calculate

avector

formedbytheanglebetweenthecorneaandpupil

reflectionsthedirectionofthisvector,combined

withothergeometricalfeaturesofthereflections,

will thenbeused to calculate thegazedirection.

TheTobiiEyeTrackers areanimprovedversion of

the traditional PCCR remote eye tracking

technology(USPatentUS7,572,008).Nearinfrared

illumination is used to create the reflection

patternson the corneaandpupilof theeyeofa

user and two image sensors areused to capture

imagesof

the

eyes

and

the

reflection

patterns.

Advanced image processing algorithms and a

Figure4PupilCentreCornealReflectiontechnique(PCCR)Alightsourceisusedtocausereflectionpatternsonthecornea

andpupilofthetestperson.Acamerawillthenbeusedto

captureanimageoftheeye.Thedirectionofthegazeisthen

calculatedusingtheanglesanddistances


9/14



www.tobii.com

7

physiological3Dmodeloftheeyearethenusedto

estimate thepositionof theeye inspaceand the

pointofgazewithhighaccuracy.

2.2 WhatareDarkandBrightPupileyetracking?

There are two different illumination setups that

canbeusedwithPCCReye tracking:brightpupil

eye tracking,wherean illuminator isplacedclose

to the optical axis of the imaging device, which

causesthepupil toappear litup(this isthesame

phenomenonthatcausesredeyesinphotos);and

dark pupil eye tracking where the illuminator is

placedawayfromtheopticalaxiscausingthepupil

toappeardarkerthantheiris.

Thereare

different

factors

that

can

affect

the

pupil

detectionduring remoteeye trackingwhenusing

each one of these two techniques. For example,

whenusing thebrightpupilmethod, factors that

affect the size of the pupil, such as age and

environmental light, may have an impact on

trackability of the eye. Ethnicity is also another

factorthataffects thebright/darkpupilresponse:

For Hispanics and Caucasians the bright pupil

method works very well. However, the method

has

proven

to

be

less

suitable

when

eye

tracking

Asians forwhom thedarkpupilmethodprovides

bettertrackability.

TobiiEyeTrackersoftheT/XSeriesusebothbright

and dark pupil methods to calculate the gaze

positionwhiletheearlier50seriesonlyuseddark

pupileyetracking.Hence,theTobiiT/XSeriesEye

Trackersareable todealwith largervariations in

experimentalconditionsandethnicitythananeye

trackerusingonlyoneofthetechniquesdescribed

above.All

participants

are

initially

subjected

to

both the bright and dark pupilmethods and the

method that is found to provide the highest

accuracy ischosenfortheactualtesting.Duringa

recording theTobiiTXSeriesEyeTrackersdonot

change between bright and dark pupil tracking

unless conditions change in a way that have a

significantlynegativeimpactontrackability.Ifthat

happens, the Tobii Eye Trackers conduct a new

testwherebothmethodsareusedsimultaneouslyinorder todeterminewhichmethod is themost

suitablefor

the

new

conditions

and

continue

the

recordingusingonlytheselectedmethod.

2.3 Whathappensduringthecalibration?

Before an eye tracking recording is started, the

user is taken through a calibration procedure.

During this procedure, the eye trackermeasures

characteristics of the users eyes and uses them

together with an internal, physiological 3D eye

model to calculate the gaze data. This model

includesinformationaboutshapes,lightrefraction

and reflectionpropertiesof thedifferentpartsof

the eyes (e.g. cornea, placement of the fovea,

etc.).During the calibration the user is asked to

lookatspecificpointsonthescreen,alsoknownas

calibrationdots.Duringthisperiodseveralimages

of the eyes are collected and analyzed. The

resultinginformation

is

then

integrated

in

the

eye

modelandthegazepointforeachimagesampleis

calculated. When the procedure is finished the

quality of the calibration is illustrated by green

lines of varying length. The length of each line

representstheoffsetbetweeneachsampledgaze

pointandthecenterofthecalibrationdot.

Large offsets (long green lines, Figure 5) can be

caused by various factors such as, the user not

actually focusing on the point, the user being

distractedduringthecalibrationortheeyetracker

notbeingsetupcorrectly.However,theuserdoes

nothave tokeep theheadcompletelystillduring

calibrationaslongasthefocusoftheuserseyesis

kepton

the

moving

dots.

During

the

calibration

both the lightanddarkpupilmethodsare tested

Figure5 CalibrationresultsEach

collected

data

point

is

compared

against

the

point

on

the

screentheuserwasaskedtolookedat.Thelinesshowthe

offsetbetweenthepointsandrespectivethegazepoints.


10/14



www.tobii.com8

to identify themost suitable for the current light

conditionsandtheuserseyecharacteristics.

2.4 Howarefixationsdefinedwhenanalyzingeyetrackingdata?

During a recording the Tobii T/X Series Eye

Trackers collect raw eye movement data points

every 16.6 or 8.3 ms (depending whether the

sampling data rate is 60Hz or 120Hz). Each data

pointwillbe identifiedbya timestampand x,y

coordinates, and sent to the analysis application

(e.g.TobiiStudiooranapplicationusingtheTobii

SDK APIs) database running on the computer

connectedtotheeyetracker.Inordertovisualize

thedata

these

coordinates

will

then

be

processed

further into fixations and overlaid on a video

recordingofthestimuliusedinthetest.

By aggregating data points into fixations the

amountofeyetrackingdatatoprocessisreduced

significantly. TobiiStudiousestwotypesoffixation

filters togroup the rawdata into fixations.These

filters are composed of algorithms that calculate

whether raw data points belong to the same

fixation or not. The basic idea behind these

algorithmsis

that

iftwo

gaze

points

are

within

a

predefined minimum distance from each other

thentheyshouldbeallocatedtothesamefixation

inotherwordstheuserhasnotmovedtheeyes

between the two sampling points. In the Clear

Viewfixationfilter it isalsopossibletosetatime

limittotheminimumlengthofafixation.Another

function of the filters is to check if the sample

pointsarevalid,e.g.discardingthepointswithno

eye position data or where the system has only

recordedoneeyeandfailedtoidentifywhetherit

is the left or the right eye and is unable to

estimatethefinalgazepoint.

Graphically fixations are typically represented by

dots (largerdots indicate a longer fixation time),

whereas saccadesare indicatedby linesbetween

fixations.Ascreenshotshowingallthefixationsa

personmade on a specific image orwebpage is

typicallycalledagazeplot(seeFigure6).Another

popular way to visualize eye tracking data is

through a heatmap (see Figure 7).Aheatmap

uses different colors to show the amount of

fixationsparticipantsmade incertainareasofthe

image or for how long they fixated within that

area. Red usually indicates a highest number of

fixationsor the longest timeandgreen the least,

with varying levels in between. An areawith no

coloronaheatmapsignifiesthattheparticipants

didnot

fixate

in

the

area.

This

does

not

necessarily

meantheydidnotseeanythinginthere,butifit

wasdetected itmayhavebeenintheirperipheral

vision,whichmeansthatitwasmoreblurred.

Figure6GazePlotorScanpathimageTheGazePlotvisualizationshowsthemovementsequenceandpositionoffixations(dots)andsaccades(lines)ontheobservedimageorvisualscene.Thesizeofthedotsindicatesthefixationdurationwhereasthenumberinthedotsrepresentstheorderofthefixation.GazePlotscanbeusedtoillustratethegazeactivityofasingletestparticipantoverthewholeeyetrackingsession,orseveralparticipantsinashorttimeinterval.

Figure7HeatmapTheHeatmapvisualizationhighlightstheareasoftheimagewheretheparticipants fixated.Warmcolorsindicateareaswheretheparticipants eitherfixatedforalongtimeoratmanyoccasions.Heatmapscanbeusedtoillustratethecombinedgazeactivityofseveralparticipantsonanimageorwebpage2.5 Ispupilsizecalculation

possiblewithTobiiEyeTrackers?

Knowing the size of the pupil and whether it

changesover

time

is

often

used

when

studying

emotional responses to stimuli. The eye model


11/14



www.tobii.com

9

usedbyTobiiEyeTrackersallows for calculations

ofthepositionoftheeyesaswellasthepupilsize.

Theopticalsensor registersan imageof theeyes

whichthen isusedtocalculatetheeyemodel.As

theeyemodelusedbyTobiiEyeTrackersprovides

dataabout

the

distance

between

the

eye

and

the

sensor,thefirmwarecancalculatethepupilsizeby

measuringthediameterofthepupilontheimage

andmultiplyitwithascalingfactor.

Severaldefinitionsexistregardingwhatshouldbe

definedas thesizeof thepupil. In theeyemodel

usedbyTobiiEyeTrackersthepupilsizeisdefined

as the actual, external physical size of the pupil.

However,inmostscientificresearchtheactualsize

ofthepupil is less importantthan itsvariations in

sizeover

time.

The

T/X

series

Eye

Trackers

outputs

pupil size information foreacheye togetherwith

eachgazepointallowinganexternalsoftware(e.g.

Tobii Studio) to record the pupil size variation

duringaneyetrackingsession.

2.6 Howdoesblinkingaffecteyetracking?

Blinking is most often an involuntary act of

shutting and opening the eyelids. During each

blink

the

eyelid

blocks

the

pupil

and

cornea

from

theilluminatorresultinginrawdatapointsmissing

the x,y coordinates information. During analysis

fixationfilterscanbeusedtoremovethesepoints

andextrapolatethedatacorrectlyintofixations.

Providedthattheheadmovementsarewithinthe

eye tracker specifications, i.e. that the missing

datapointsdonotoriginatefrommovingthehead

awayfromtheeyetrackingbox1,itisalsopossible

toextractinformationonblinksfromtherawdata

collectedby

the

eye

tracker.

This

can

be

done

by

extractingitmanuallyfromtherawdataexported

fromTobiiStudio.

1An eye tracking box is the area in front of an eye tracker

withinwhichtheusercanmovewithouttheeyetrackerlosing

theability to track theeyes.TobiiEyeTrackershavedifferent

eye

tracking

boxes

dependent

on

model:

For

trackers

running

in60HztheboxmeasuresaboutW:44cmxH:22cmat70cm

fromtheeyetrackerand for120HzeyetrackersW:30xH:22

cmat70cmdistancefromtheeyetracker.

2.7 Doesheadmovementaffecteyetrackingresults?

During an eye tracking session headmovements

withintheeyetrackingboxhaveverylittleimpact

on the gazedata accuracy. The optical sensorof

theTobiiEyeTrackersiscomposedoftwocameras

that capture an image of the eyes at a given

frequency (60 Hz or 120 Hz). The two cameras

produce two images of the eyes simultaneously

and the respective pupil and corneal reflections

providing the eye tracker with two different

sourcesofinformationregardingtheeyeposition.

This type of stereo data processing offers a

robust calculation of the position of the eye in

spaceandthepointofgazeevenifthepositionof

thehead

changes.

Additionally the Physiological 3D Eye Model of

eachparticipantseyeoffersanaccurateandmore

robustway todetermine thepositionof the eye

andpointofgazeof theparticipant independent

ofheadmovement.

3 WhatinfluencestheaccuracyofaTobiiEyeTracker?

There are several factors that can affect the

accuracyofeye trackingresults;among themare

eye movements, the calibration procedure, drift

andambient light.Eye trackers suchasTobiiEye

Trackers can record saccades and fixations, pupil

sizeandotherusefuldatawithahighaccuracy.

3.1 EyemovementsAsmentionedat thestartof thiswhitepaperwe

perceiveimagesorscenesbymovingoureyesand

fixatingon

areas

of

interest

and

that

during

each

fixationweareessentiallyplacingourfoveaonthe

area or feature we wish to extract more detail

about.Thesizeofthefoveavariesbetween12of

the visual field meaning that if we stand at a

certaindistance fromanobject, theareacovered

bythefoveawillbeaprojectionofthesizeofthe

fovea ina12degreeangle (seeFigure8on the

nextpage).Whenwemove the fovea inorder to

place itonareasweare interested inwedonot

needtoplaceitexactlycenteredandontopofthe

areaas

the

projected

area

becomes

larger

and

hence,coversmore,thefurtherawayanobjectis.


12/14



www.tobii.com

10

Infact it isquitecommonthatthefoveaoverlaps

partially on the area and that this is enough to

extractthe levelofdetailweneed, inparticular if

we are looking at a familiar scene or image. In

addition,humans

are

generally

unable

to

voluntarily direct their eyes to very precise

locations fora longperiodof time.Duringsteady

fixations the human eye is in constant motion;

small involuntary movements are triggered to

avoid perceptual fading (overstimulation of the

lightreceptorsthatcausestheneuronstoceaseto

respondtothestimulus).Thus,evenifweperceive

thatweare lookingata specific spotona scene

our eyes are actually moving between different

locations

around

the

spot.

Even

though

the

accuracyofeye trackingresultsare influencedby

human vision accuracy limitations, this influence

only surfaces when doing very fine grained

accurate studies. An eye tracker can record and

replaywhatpeoplearelookingatwithin lessthan

acentimetersaccuracywhenmeasuringthepoint

ofgaze

on

asurface

or

screen

under

normal

test

conditions.

3.2 DriftDriftisthegradualdecreaseinaccuracyoftheeye

trackingdata compared to the trueeyeposition.

Drift can be caused by different factors, such as

variations in eye physiology (e.g. degree of

wetness,tears)andvariations intheenvironment

(e.g.sunlightvariations).However,driftproblems

only

have

a

significant

effect

if

the

test

conditions

change radically or if the eye tracking session is

long. In these cases it can be attenuated by

Figure8Calculatingthesizeofthevisualangleonascreen.Inthisexampleweassumethatthefoveaisapproximately2ofvisualangleandwewanttomeasurethecorresponding sizeofthefovea

onthescreenwhenstandingatthedistanceof64.Tocalculatethesizeoftheprojectionwecanusebasictrigonometry(seeinformation

boxonthefigure):firstweneedtoassumethatwecandividetheprojectionofthefoveaintotwoequalrightangledtriangles(mirroring

eachother).Thustheanglethatwewilluseintheformulawillbeequaltothevisualangledividedby2.Secondlywemeasurethedistance

tothescreen.Wewillusethatvalueasthemeasurementofthehypotenuseofthetriangle(adjacentside). Thirdly,weapplythetwo

values(angle/2andthedistancetothescreen)totheformulaandobtainthemeasurefortheshortestsideofthetriangle(oppositeside).

Finally,toobtainthevalueforthesizeoftheprojectionofthevisualangleonthescreenwemultiplythevaluefortheshortestsideofthe

trianglebytwo.


13/14



www.tobii.com

11

recalibratingfrequently.Today,manyeyetrackers

(including the Tobii T/X series) are able to cope

wellwith drift, however extreme changes in the

eyephysiologyduringaneye trackingsessioncan

stillproduceasignificantdrifteffect.

4 Whatdoeseyetrackingdatatellus?

Eye tracking analysis is based on the important

assumption that there is a relationship between

fixations, our gaze and what we are thinking

about.However,thereareafewfactorsthatneed

to be considered for this assumption to be true

whichwillbediscussedbelow.

First, sometimes fixations do not necessarily

translate into a conscious cognitive process. For

example,duringasearchtaskonecaneasilyfixate

brieflyonthesearchobjectandmissitspresence,

especiallyiftheobjecthasanunexpectedshapeor

size (commonly called change blindness). This

happens because our expectation of what the

object (or scene) should look likemodulates our

visual attention and interferes with the object

detection. This effect can be eliminated from a

testifyougiveclearinstructionstotheparticipant,

and/orfollow

up

the

eye

tracking

test

with

an

interviewtoassesstheparticipantsmotivationsor

expectations.

Second, fixations can be interpreted in different

ways depending on the context and objective of

thestudy.Forexample,ifyouinstructaparticipant

to freely browse a website (encoding task), a

higher number of fixations on an area of the

webpage may indicate that the participant is

interested in that area (e.g. a photograph or a

headline)or

that

the

target

area

is

complex

and

hard to encode. However, if you give the

participantaspecificsearch task (e.g.buyabook

on Amazon), a higher number of fixations are

often indicative of confusion and uncertainty in

recognizing the elements necessary to complete

the task. Again, a clear understanding of the

objectiveof thestudyandcarefulplanningof the

tests are important for the interpretation of the

eyetrackingresults.

Andthird,

during

the

processing

of

avisual

scene,

individuals will move their eyes to relevant

featuresinthatscene.Someofthesefeaturesare

primarily detected by the peripheral area of our

visual field. Due to the low acuity, a feature

located in thisareawill lackshapeorcolordetail

butwe are still able to use it to recognizewell

knownstructures

and

forms

as

well

as

make

quick,

general shape comparisons. As a result, we are

able touse theperipheralvision to filter features

according totheirrelevancetous,forexample, if

we generally avoid advertisement banners on

webpages,wemightalsoavoidmovingoureyesto

othersectionsofthewebpagethathaveasimilar

shape simplydue to the fact thatourperipheral

vision tellsus that theymightbebanners. The

currenteyetracker technologywillonlyshowthe

areasonthevisualscenethatthetestsubjecthas

beenfixatingatandthejumpsbetweenthem(i.e.

not the whole visual field). Thus, to fully

understandwhyatestpersonhasbeenfixatingon

some areas and ignoring others, it is important

that the tests should be accompanied by some

formofintervieworthinkaloudprotocols.

5 ConclusionEyetracking isan importanttechniquethatoffers

anobjective

way

to

see

where

in

ascene

a

persons visual attention is located. However, as

withanyotheranalyticaltechnique,itisnecessary

to havea clearmethodology that is adequate to

thecontextandobjectivesofthestudyifwewish

tounderstandand interprettheeyetrackingdata

correctly.


14/14


Tobii Technology in the United States and/or other countries.12

6 BibliographicsourcesAnderson J.R. 1995. Visual attention. In Cognitive Psychology

and its Implications.Ed.WHFreeman&CompanyNewYork,

4thEdition,pp.81105.

Busey T.A. & Loftus G.R. 1994. Sensory and cognitive

components of visual information acquisition. Psychology

Review.pp.446469.

EhmkeC.&WilsonS.2007.IdentifyingWebUsabilityProblems

fromEyeTrackingData. InPeopleandComputersXXIHCI...

butnotaswe know it:ProceedingsofHCI2007.Ed.Ball L.J.,

SasseM.A.,SasC.,OrmerodA.,DixA.,BagnallP.& Ewan T.,

BritishComputerSociety,pp.109116.

Lee J.C.,Kim J.E., ParkK.M. &Khang,G.2004.Evaluationof

themethods for pupil size estimation: on the perspective of

autonomic activity: Engineering in Medicine and Biology

Society,2004.

IEMBS

'04.

26th

Annual

International

Conference

oftheIEEE,SanFrancisco,CA, pp.15011504

Pool A. & Ball L.J. 2006. Eye Tracking in HumanComputer

InteractionandUsabilityResearch: Current Status and Future

Prospects.InEncyclopediaofHumanComputerInteraction.Ed.

C.Ghaoui, IdeaGroup,Inc.Pennsylvania,pp.211 219.

Rayner K.1998. EyeMovements inReadingand Information

Processing: 20 Years of Research. Psychological Bulletin, 124,

pp.372422.

Rayner K.2009. EyeMovements andVisualEncodingDuring

ScenePerception.PsychologicalScience,Vol.20,Nr1,pp.610

SpakovO.2008iComponent DeviceIndependentPlatformfor

Analyzing Eye Movement Data and Developing EyeBased

Applications. Doctoral dissertation. Acta Electronica

UniversitatisTamperensis,725.

Date post:	30-May-2018
Category:	Documents
Upload:	jon-ward
View:	221 times
Download:	0 times

Introduction to Eye Tracking and Tobii Eye Trackers

Documents