Behavior Research Methods&Instrumentation1975, Vol. 7 (5),397-429

METHODS & DESIGNS

Survey of eye movement recording methods

LAURENCER. YOUNGDepartment of Aeronautics and Astronautics, Massachusetts Institute of Technology

Cambridge, Massachusetts 02139

and

DAVID SHEENAWhittaker Corporation, Space Sciences Division, 335 Bear Hill Road, Waltham, Massachusetts 02154

This paper reviews most of the known techniques for measuring eye movements, explaining theirprinciple of operation and their primary advantages and disadvantages. The five sections of the papercover the followingtopics: (1) types of eye movement, (2) characteristics of the eye which lend themselvesto measurement and the principal approaches to the measurement of eye movement, (3) practical methodsof measurement with especial attention to the new techniques, (4) general considerations guiding aselection of method, and (5) summarizing of the major findings in a concise table.

TYPES OF EYE MOVEMENTl

For any application of eye movement instrumenta­tion, the kinds of eye movements to be observed mustbe clearly understood so that the instrument specifica­tions may be properly assigned. The following list sum­marizes the types of known eye movements. Eye move­ments are considered here as rotations about a horizon­tal axis, an (initially vertical) axis which rotates with theglobe about the horizontal, and a torsion axis alongthe angle of gaze. The eye does not rotate about a fixedcenter, but this is not of practical importance for mostmeasurement applications (Park & Park, 1933). Severaltypes of horizontal eye movements are illustratedin Figure 1.

Saccadic eye movements are the rapid conjugatemovements by which we change fixation from one pointto another voluntarily. They include the "jump andrest" fixation movements observed in scanning a visualscene or reading. They are characterized by very highinitial acceleration and final deceleration (up to40,000 deg/sec") and a peak velocity during the motionwhich varies with the amplitude of the saccade and maybe as high as 400 to 600 deg/sec, The duration of asaccadic eye movement also varies with its magnitudefrom 30 to 120 msec (Mackensen, 1958). Saccadiceye movements generally observed in searching are ofthe range of 1 to 40 deg. Head motion is often involvedwhen the target motion exceeds 30 deg. In response toa visual stimulus, saccadic eye movements exhibit a

Prepared for Conferences on Eve Movement Research andTechnology, Task Force on Essential Skills. National Institute ofEducation.

latency of 100 to 300 msec. Vertical or oblique saccadiceye movements may have a torsional component associ­ated with them because of the arrangement of the sixextraocular muscles. The purpose of the saccadic eyemovement system appears to be fixation of the imageof the target on the fovea, or high-acuity region of theretina, corresponding to .6 to 1 deg of visual angle.There is a minimum delay or refractory period betweensaccadic eye movements of 100 to 200 msec. The visualthreshold is significantly elevated during the period justprior to and during a saccadic eye movement (Cook,1965; Robinson, 1964: Vossius, 1960; Young, Zuber, &Stark, Note 1).

Pursuit, or slow-tracking, movements are conjugateeye movements used to track slowly moving visual tar­gets in the range of 1 to 30 degjsec. Pursuit movementsare smoothly graded and appear to partially stabilize theimage of the moving target or background on the retina,independent of the saccadic eye movement system. Thepursuit-movement system appears to be limited in accel­eration as well as in velocity. Smooth pursuit movementsare not generally under voluntary control and usuallyrequire the existence of a moving visual field for theirexecution (Robinson, 1965; Westheimer, 1954).

Compensatory eye movements are smooth move­ments, closely related to pursuit movements, whichcompensate for active or passive motion of the head ortrunk. They tend to stabilize the retinal image of fixedobjects during head motion and are attributable both tosemicircular-canal stimulation sensing head motion andto neck proprioception associated with the turning ofthe head on the trunk. Compensatory eye movementsare closely related to the slow-phase eye movements ofvestibular nystagmus discussed below.

397

398 YOUNG AND SHEENA

1-1-1-1-+-1-1--1-1--1- -I 1 1-1-' 1-' 1 1- -I " ill t--J· i ' I iI' , I i 1_

1-1--1---1--1 -1--; -1-1--1--; ; I ! 1-;-1 1 , I I I , I I t 1-t-I--+--t -I '-r-+-t-t- 1-- t- 1-1--1 +-1 1 , I

I

iI

i I i-I I 1 I

! i I I ! ! I

: : ~ '. '. ~ ! ; ;"~--T--r~H--t-t-:-

I '1 ••c. ,; SCALES, 20 mm/ase

1.4°/div

Figure 1. Typical horiz~nt~ eye movements recorded with a photoelectric monitor showing saccadic jumps, fixation movements,smooth pursuit, and optokinetic nystagmus. From Young, 1970 (Copyright 1970, McGraw-Hill Book Company. Used with permis­sion of McGraw-Hill Book Company).

Vergence eye movements are movements of the twoeyes in opposite directions in order to fuse the image ofnear or far objects. The vergence movements are consid­erably slower and smoother than conjugate eye move­ments, appear to be nonpredictive, and reach maximumvelocities on the order of 10 deg/sec over a range ofnearly 15 deg. They are stimulated by focusing error(accommodative convergence) as well as binoculardisparity (Rashbass & Westheimer, 1961; Zuber & Stark,1968).

Miniature eye movements, or fixation movements,include a variety of motions which are generally lessthan 1 deg in amplitude and occur during attemptedsteady fixation on a target. Drift is the slow randommotion of the eye away from a fixation point at veloci­ties of only a few minutes of arc per second occurringwithin the foveal dead zone. Flicks, or microsaccades,are small rapid eye movements which have been shownto be dynamically of the same nature as large voluntarysaccades, of magnitudes as large as 1 deg, occurring atintervals separated by as little as 30 msec and whichgenerally redirect the eye toward the position necessaryfor fixation on the target (Zuber, 1965). Both flicks anddrifts tend to occur along a single preferred axis in anyindividual (Nachmias, 1959). There is currently no gen­eral agreement on the error-correcting nature of theflicks or drifts. In addition, normal individuals fixatingon targets exhibit a high-frequency tremor in the rangeof 30 to 150 Hz with peak amplitudes of approximately30 arc sec in the region of 70 Hz. (Because of thepresence of these fixation movements, accuracy of .5 to1 deg is often sufficient in eye-monitoring tasks designedto show what part of the visual field is being fixated.)

Optokinetic nystagmus, also known as "train nystag­mus," is a characteristic sawtooth pattern of eyemotion elicited by a moving visual field containing re­peated patterns. Optokinetic nystagmus consists of aslow phase in which the eye fixates on a portion of themoving field and follows it with pursuit motion and afast phase or return saccadic jump in which the eyefixates on a new portion of the field. The minimum timebetween fast phases is approximately .2 sec, resulting in

a maximum frequency of approximately 5 Hz, althoughthe nystagmus frequency may be considerably less forslow field motions. The amplitude of optokinetic nystag­mus is variable, generally from 1 to 10 deg. If a nonmov­ing fixation point is present in the visual field, thenystagmus response may be reduced to a fraction of adegree, which is not noticeable by direct observation.Attempts have been made to use optokinetic nystagmusas an objective measurement of visual acuity by deter­mining the minimum line width which induces nystag­mus.

Figure 1 shows typical single-eyed horizontal eyemovements recorded by the photoelectric technique.The rapid saccadic jumps and fixation movements areseen in the early part of the trace, followed by smoothpursuit movements in tracking a target, and finally opto­kinetic nystagmus induced by horizontal movementof a sheet oflined paper.

Vestibular nystagmus is an oscillatory motion of theeye, similar in appearance to optokinetic nystagmus,containing a slow phase and a fast saccadic-like return.It is primarily attributable to stimulation of the semicir­cular canals during rotation of the head with respect toinertial space. A counterclockwise head rotation abouta vertical axis leads to deflection of the cupulas of thehorizontal semicircular canals, which induces image­stabilizing slow-phase eye movement in a clockwisedirection. As head motion continues, the eyes jump backrapidly to pick up another position and repeat the saw­tooth pattern. Measurement of vestibular nystagmus invarious axes is a commonly used test of the semicircular­canal function, either through the threshold of angularacceleration impulse required to induce nystagmus orthrough the duration of "postrotation nystagmus." Thelatter test constitutes a part of the nystagmus or objec­tive cupulogram. The amplitude and frequency of vestib­ular nystagmus are similar to those of optokineticnystagmus. It has recently been demonstrated that vesti­bular nystagmus can be induced by pure linear accelera­tion and by a rotating linear-acceleration vector; theformer may reflect otolith contribution, while the latteris attributable to deformation of the semicircular canals

SURVEY OF EYE MOVEMENT RECORDING METHODS 399

(Steer, 1967; Young, 1969). A variety of rotationaltests are used with eye movement measurements toindicate the pathology of the nonauditory labyrinth. Arelated type of eye movement induced in the clinical sit­uation is caloric nystagmus. The external semicircularcanals are stimulated by the convection currents inducedin the canals upon introduction of water above or belowbody temperature into the external ear. The temperaturechange necessary to induce nystagmus and the durationand strength of nystagmus for standard temperaturesconstitute a set of clinical measurements useful in thediagnosis of vestibular disease.

Spontaneous nystagmus, or gaze nystagmus, is ananomaly of nystagmus associated with a number ofneurological disorders. This nystagmus may be eitherlarge enough for direct observation or less than 1 deg,requiring recording for detection. Gaze nystagmus isobserved only when the subject looks in a certaindirection. The nystagmus may be either asymmetric,containing a slow phase and a fast phase, or "pendular,"showing fine high-frequency oscillations of from 4 toI0 Hz. Some of these abnormal eye movements are toosmall to be observed in close clinical examination andare important diagnostically. Recording of nystagmus aswell as the ability of subjects to follow targets withnormal saccadic and pursuit tracking has become wide­spread.

Torsional eye movements are rotation movements ofthe eye about the line of gaze, and they are generallylimited to angles of less than 10 deg. The rolling motionsmay be stimulated by rotational optokinetic nystagmusor by two types of vestibular responses. The torsionalcomponent of vestibular nystagmus or compensatoryeye movement in response to head rotation is similar tothe horizontal and vertical vestibular nystagmusmentioned above. In addition, the phenomenon ofcounterrolling or steady state offset of the eye about thetorsional axis in response to tilt of the head with respectto the vertical, has been demonstrated and shown to beattributable to the human otolith or vestibular gravityreceptors. As such, it has been suggested as an importantmeasurement of otolith function (Woellner & Graybiel,1959).

PHYSICAL CHARACTERISTICS OF THE EYEWHICH ARE USED IN

EYE MOVEMENT MEASUREMENT

The RetinaThe eye has no proprioceptive feedback, in terms of

conscious position sense. It does, however, contain theretina which moves with the eye and makes possiblethe subjective assessment of eye movement. Among theearliest quantitative techniques for determining thevelocity of the eye during pursuit and saccadic eyemovements was after-images, A small light, flashed

periodically, will leave a trace of after-images, thedensity of which indicates fixation duration and thespacing of which indicates the velocity of eye move­ments. After-images separated by as little as 15 arc mincan be resolved, and the technique is usable over theentire range of eye movements. Its chief drawbacksare, of course, the subjective nature of the measurementand the fact that it can only be used for a brief interval,after which the subject must report on the numberand placement of his after-images. It is of practical usecurrently only for the measurement of ocular torsion,where it provides a convenient and relatively accuratemeasurement and for which there are no readilyavailable automatic methods which are economical,simple to apply, and easily analyzed.

The fovea contains thousands of light-absorbingradially oriented crystals which selectively absorblinearly polarized light. Kaufman and Richards(Kaufman & Richards, 1969; Richards & Kaufman,1969) used rotating polarized blue light to form a "spin­ning propeller" after-image relative to the subject'sfovea to determine fixation points.

Corneo-Retinal PotentialA potential difference of up to 1 mV between the

cornea and the retina (cornea positive) normally existsin the eye and is used as the basis of the most widelyapplied clinical eye movement technique-electro­oculography. The precise basis of this potential differ­ence, once attributable to the electrical activity of theretina itself, is now in question once again. This poten­tial has important variations diurnally and also withthe level of light adaptation, decreasing following steadyperiods in the dark (Gonshoor & Malcolm, 1971).For stable electro-oculographic measurements, especiallyin the dark, the subject should be permitted to adaptto the ambient illumination level to be used in theexperiment for 30 to 60 min prior to the experiment.

The negative electrical pole lies approximately at theoptic disk, 15 deg displaced from the macula. Since theelectric field is not aligned with the optic axis, anytorsional rotation of the eye introduces a potentialchange which can be mistaken for horizontal or verticaleye movement. This very geometry, however, makeselectro-oculography a possible, though difficult, methodfor measuring ocular torsion (Gabersek, 1969).

Electrical ImpedanceThe impedance measured between electrodes placed

at the outer canthi of the two eyes varies with eyeposition. The variation in this resistive component iseither associated with the nonhomogeneous or aniso­tropic nature of electrical characteristics of the tissuesin the globe or with the nonspherical characteristics ofthe globe so that the resistivity of the path between thetwo electrodes changes with position (Geddes, McGrady,& Hoff, 1965; Sullivan & Weltman, 1963).

400 YOUNGAND SHEENA

The Corneal BulgeThe cornea, attached to the sclera at the front of the

eye and centered close to the optic axis, has a smallerradius of curvature than the eye itself. This forms thebasis for a number of important methods of eye move­ment measurement. In the early days of research on eyemovements, attachments were made directly to thecornea by a plaster of paris ring and mechanical linkagesto recording pens (Delabarre, 1898). The bulge of thecornea can be felt through the eyelid of the closed eye,and pressure transducers placed over the eyelid candetect these changes. In more recent times, the corneahas acted as a mechanical post to center tight fittingscleral contact lenses to which other measurementdevices are attached. It should be noted that the corneaitself slips slightly with respect to the sclera when forcesare applied to the cornea and probably slips slightlyduring the eye acceleration phase of saccadic eye move­ments. Contact lenses applied to the cornea itselfare not an adequate base for the accurate measurementof eye position, and large contact lenses conforming tothe sclera as well as the cornea are necessaryfor systems in which stability of better than a fewminutes of arc is desired. The nominal curvature of thecornea for an adult human is approximately an 8-mmradius for an eye of 13.3-mm radius. Once a contactlens is fitted, its position can be measured by any of anumber of methods.

Corneal ReflectionsThe front surface of the cornea, although not a per­

fect optical surface, approximates a spherical sectionover its central 25 deg. As with a convex mirror, reflec­tions of a bright object from this surface form a virtualimage behind the surface which can be imaged andphotographed or recorded. The position of the cornealreflection, commonly seen as the highlight in the eye,is a function of eye position. Rotation of the eye aboutits center produces a relative translation, as well asrotation of the cornea, forming the bases for the impor­tant class of eye movement instruments known ascorneal reflection systems.

Reflections from Other Optical Curvatures in the Eye­Purkinje Images

Although the brightest reflections of incident lightcome from the front surface of the cornea, light is alsoreflected from each surface of the eye at which there isa change in refractive index. Reflections come also fromthe back surface of the cornea, the front surface of thelens, and the rear surface of the lens. These four arereferred to as the Purkinje Images. After the brightfront surface reflection, the next most visible Purkinjeimage is the fourth, coming from the posterior surfaceof the lens. Measurements of the relative displacementbetween the first and fourth images, representing, as

they do, points focused and imaged from planes of dif­ferent depths in the eye, represent one technique foractively measuring the orientation of the eye in spaceindependent of its relation to head position.

The LimbusThe iris of the eye is normally visible and clearly dis­

tinguishable from the sclera and is the. basis for thenormal visual assessment of the angle of gaze. Theposition of the iris-scleral boundary (the limbus) maybe measured with respect to the head. The ratio betweendark iris and bright sclera observed on the left and rightsides of the eye may either be measured directly withphotosensors or indirectly on an image of the eye. Thisratio is directly related to the horizontal position of theeye. The best wavelength for making the distinctionbetween iris and sclera depends to some extent on theiris color; however, white light is normally reasonablyeffective.

The PupilThe pupil is easily distinguished from the surrounding

iris by its difference in reflectance. The pupil can bemade to appear much darker than the iris under mostlighting conditions when the majority of light does notcome in directly along the axis of measurement and,consequently, is not reflected out. On the other hand,the pupil can be made to appear very bright (as oftenseen in amateur full-face flash photography) when mostof the light enters along the optic axis and is reflectedback from the retina. In either case, the pupil can beseparated from the surrounding iris optically. This canbe especially sharpened with the use of infrared lightwhich will be nearly entirely absorbed once enteringthe eye, consequently making the pupil much darkerthan the surrounding iris. The pupil normally varies be­tween 2 and 8 mm in diam in adult humans. Althoughit is actually slightly elliptical in shape, it can be approx­imated closely by tracing the best-fitting circle to thepupil circumference with an image dissector technique.The center of the pupil is also easily located electro­optically or on film for hand analysis.

The pupil appears elliptical when viewed other thanalong the optic axis, with the minor axis shortening asthe eye rotates. The pupil eccentricity could serve as abasis for eye angle measurement.

Other Optical and Nonoptical LandmarksIn addition to the iris and the pupil, other optical

landmarks can be traced. Scleral blood vessels or folds ofthe iris can be identified by hand or traced ~th opticaltracing techniques. (These are of practical applicationonly in the measurement of ocular counterrolling.) Theretinal vessels, which can also be imaged and tracked,provide one of the most accurate techniques for deter-

SURVEY OF EYE MOVEMENT RECORDING METHODS 401

mining the place on the retina where a given target isimaged and consequently the exact fixation point of theeye. The retinal vessels, approximately .2 mm in diam,radiate from the optic disk.

Some artificial landmarks have also been placed onthe eye and their positions recorded. A globule ofmercury (Barlow, 1952), chalk, and egg membranehave been used for optical tracking. A small piece ofmetal imbedded in the sclera has also been used for mag­netic tracking of the eye position.

MAJOR EYE MOVEMENTMEASUREMENT TECHNIQUES

Electro-OculographyPrinciple. Mowrer, Ruch, and Miller (1936) and

Schott (1922) found that the position of the eye couldbe measured by placing skin electrodes around the eyeand recording potential differences. The source of theelectrical energy is the corneoretinal potential, orelectrostatic field which rotates with respect to the eye.The cornea remains .40 to 1.0 mV positive with respectto the retina; this is attributable to the higher metabolicrate at the retina. As the eye rotates, the electrostaticdipole rotates with it, and consequently the potentialdifference in a plane normal to the principal axis varies,theoretically, as the sine of the angle of deviation. Ofcourse, the nonhomogeneous nature of the conductingmedium causes wide departure from the theoreticalvalues. These techniques were reviewed by Marg (1951)and more recently by Kris (1960) and by Shackel(1967). Skin electrodes placed on the outer canthimeasure conjugate horizontal eye position. By referenceto any third electrode over the bridge of the nose, somemeasure of horizontal eye vergence may also bedetected. The recorded potentials are small, in the rangeof 15 to 200 microV, with nominal sensitivities of theorder of 20 microV/deg of eye movement (Shackel,1967).

Implementation. The signals are at times difficult todetect in the presence of large muscle-action-potentialartifacts, also picked up as potential differences bythe skin electrodes (Shackel, 1960). The presence ofexternal electrical interference is troublesome unless careis taken to shield the system.

Normally, but not always, the de recording method,which is necessary to determine eye position is referredto as electro-oculography (EOG); ac recording,which is useful for measurement of eye movement,including the fast and slow phases of nystagmus, is refer­red to as electronystagmography (ENG). Untilrecent years, the problem of drift in both electrodes andde amplifiers made de recording a difficult and frustrat­ing practice. Gross patterns of reading movements (num­ber of saccades per line; number of regressions and fixa­tion durations) can easily be obtained from ac recordingwith time constants of 3 to 10 sec or more. With longertime constants, even the approximate position ona line can be determined. An advantage of ac recording

+[>VerticalPosition

DC Amplifier

[>Hom ontolPositron

DC Amplifier

Figure 2. A method for reducing cross coupling in conjugateelectro-oculography (Young, 1970 after Jeannerod et al., 1966).

is that a high-sensitivity recording can be made withoutfear of having the record drift off scale.

Recently, de recording has been made much morepractical with two electronic advances. New skinelectrodes, especially silver-silver chloride (Beckman orBecton-Dickinson, for example) are easily applied withadhesive tape rings, require no skin preparation otherthan cleaning with alcohol or acetone, are of minimaldiscomfort, and resist excessive polarization over manyminutes of use. They also seem to be relatively lesssensitive to changes in skin resistance when led into anyof the newer high-input impedance (FET) preamplifiers.Gold-plated electrodes are also in use (Toglia, 1973).Placement of a high common mode rejection preampli­fier very close to the subject's head, proper grounding ofthe subject through an ear electrode, and use of shieldedcable can help eliminate the disturbance from electro­magnetic pickup.

Conjugate horizontal eye movements are recordedbetween electrodes at the outer canthi of the eyes. Place­ment of the electrodes further back toward the templesreportedly reduces the artifacts from muscle activity.Separate recording of the horizontal movements of theeye is generally performed with the use of a thirdcommon electrode placed over the bridge of the nose.

When both vertical and horizontal recording from oneor both eyes is performed, the errors introduced bycoupling between the axes and by the nonlinearity ofthe records can be considerable. Some improvement inthe cross coupling results from the use of "vectorelectro-oculography" introduced by Uenoyama,Uenoyama, and Iinuma (1964) and extended byJeannerod, Gerin, and Rougier (1966). In addition tosimultaneously displaying the x and y coordinates of eyemovements, they electrically short circuit the two super­ior electrodes and the two inferior electrodes for the ver­tical eye movement recording as seen in Figure 2. Acommonly observed overshoot artifact in EOG recording

402 YOUNG AND SHEENA

Collimated Incident Beom

Figure 3. Corneal reflection geometry.

t111111\tl

Displaced CornealReflection

rt = l a t e r a I d t ap l ar-emenr (If f>y(:' cenr e r

r .. lateral displacement of cen re r ofcurvature of cornea due to rotation

A ., radius of t he eye (from cent er ofrotation to outer central au r-Faccof cornea (...... 13.3 rrrn)

a " radius of curva r ur c of our ers u r f ace of co r nca (...... 8 nnn)

x. d + rbut r" (A-a) sin g

(x =. .J + (A-a)sin91sin b- x/a

" = Li~0:2510-1 .,,/ a . do = 2 510-] -<1 -1) sin Q + -a)

5101,.= «(~ -1) sinQ +~)for small a;~lE's. sinfl:::.-Q

sin~= (~ -1)9 ~ )

or I" '" J: (~ -l)A ~ ) I

Displaced Centerof Eye

VISiOn camera element, or photocells (Carmichael &Dearborn, 1947; Taylor, 1971; American Optical,Note 3). An incident ray is reflected at an angle 1/>,which varies with the displacement (x) of the center ofcurvature of the cornea perpendicular to the incidentray. As seen in the diagram, this displacement consistsof two parts, a displacement (r) resulting from eyeballrotation relative to the light source and a displacement(d) equal to the linear displacement of the center ofrotation of the eye normal to the incident ray. Theapparent displacement of the corneal reflection to a sta­tionary observer is x = d + (A-a)sin O. The expectedrelationship among reflection angle (I/» of a single beamof light, eye rotation (0), and lateral displacement ofthe center of the eye (d) is:

sin I/> = 21(~-1) sin 0 + ~I·a <' a

Usually the small angle approximation for 1/>, as well asfor 0, is used (Ditchburn and Ginsborg, 1953), yielding:

I/> ~ 21(~-I)O +: I,which is valid only for small eye rotations and for small

I

~dIIIIIIIIIIII

Center .1 Eye ----.... d

Reflectedalom

Corneal ReflectionPrinciple. The corneal bulge produces a virtual image

of bright lights in the visual field and region. Becausethe radius of curvature of the cornea is less than that ofthe eye, the corneal reflex moves in the direction ofeye movement, relative to the head. Since it only movesabout half as far as the eye, it is displaced opposite tothe eye movement relative to the optic axis, or thecenter of the pupil, as seen in the accompanying figures.

The geometry of the corneal reflex principle is seenby reference to the accompanying Figure 3. Incidentlight from a source (which is here shown to becollimated for illustration) is reflected from the convexsurface of the cornea in a pattern of diverging light andis imaged through a concave lens onto a film plate, tele-

Right eye horizontal: (A - C) +~(D +E)

Left eye horizontal: (C - B) + "'((F +G)

Left eye vertical: F - G

of vertical eye movements has been attributed to themotion of the upper eyelid.

An alternative combination of electrode outputshas been found useful for simultaneous separate record­ing of the vertical and horizontal movements of eacheye (Bles & Kapteyn, 1973). By combining the indi­vidual electrode outputs of Figure 2, as indicated below,one can choose ~ and "'( during calibration to virtuallyeliminate the effect of vertical eye movements:

Right eye vertical: D - E

For animal work, the stability and signal-to-noiseratio of electro-oculography can be improved consider­ably by the use of fine platinum needle electrodes in theskin around the orbit or by the use of miniature perma­nently implanted silver-silver chloride electrodes placedin holes in the bony orbit (Bond & Ho, 1970).

Evaluation of Electro-Oculography. EOG has thelargest range of any of these objective methods practicalfor human studies, since it does not require visualizationof the eye. The method is usable for eye movements upto ±70 deg (Wolf & Knemeyer, Note 2). Linearitybecomes progressively worse at excursions greater than30 deg, especially in the vertical. Typical accuracywith surface electrodes is ±1.5-2 deg.

The chief sources of error are muscle artifacts, eyelidinterferences, basic nonlinearity in the technique, andvariation in the corneoretinal potential attributable tolight adaptation, diurnal variations, and the state ofalertness.

Potential improvement in EOG performance couldcome with the making of integrated electrode amplifierassemblies attached directly to the skin to eliminatenoise susceptibility and minimize shielding requirements.

SURVEY OF EYE MOVEMENTRECORDING METHODS 403

Figure 4. The AO ophthalmograph (Taylor, 1971). Courtesyof Educational Development Laboratories, A Division ofMcGraw-Hill Book Company .

4>, meaning that the reflected beam is close to the inci­dent beam. For side lighting, with a large initial angleof reflection (4)0), the appropriate equation is:

sin(4)o + ~4» =21(~- l)sin 8 + ~I·

or

~4>?: 21(~ - 1)+1}/cos 4>0

for small eye rotations (8) and any initial reflectionangle.

The lateral head movement factor (d/a) can contri­bute a large error when the head moves relative to thelight. Ditchburn and Ginsborg point out that, withA =13.3 mm and a =8.0 mrn,

placing the light source in the target field, movements ofthe corneal reflex relative to the pupil indicate the pointof regard of the eye in the field, and not relative to thehead. These techniques are referred to as corneal reflexpoint of regard instruments and are much less sensitiveto head position. They are described in detail in a latersection.

Laboratory corneal reflex camera. In the early AOophthalmograph, and its descedent, the Reading Eye I,the head was fixed with a chin rest and head rest, andreading material was presented on fixed index cards.Small dim lights located temporally in the peripheralfield reflected corneal highlights which were imaged bya camera (See Figure 4). The horizontal motion of thecorneal reflex of each eye is recorded as a time trace onmoving film, yielding a record such as that shown inFigure 5. Vertical and horizontal scan patterns can beshown simply by stopping the film for a period. Timingmarks can be added by periodic interruptions of thelight.

As a variation of the basic corneal reflection tech­nique to permit continuous monitoring of vertical aswell as horizontal eye motion, the corneal reflex issplit into two beams by prisms. One beam is recordedon vertically moving film to yield horizontal eyemovements, and the second beam is directed to a hori­zontally moving film to record vertical movements(Buswell,1935).

One laboratory camera is presently available, thePolymetric eye movement recorder V-l164. It providesfor the superposition of the corneal reflection spoton movie film or a television picture of the scene(Mackworth & Mackworth, 1958). It provides a ±.5-degaccuracy, but strict head fixation is required. The

Figure 5. Continuous moving film record of corneal reflec­tions (Taylor, 1971)(Courtesy of Educational DevelopmentLaboratories, A Division of McGraw-HilI Book Company).

4> = 1.38 +860d,

with 4> and 8 measured in arc minutes and d in mil­limeters [l-mm change in head position is equiva­lent to an. eye rotation of greater than 12 deg.) Forthis reason, many of the fixed head versions and thehead-mounted devices require accurate stabilization ofthe light source and recording device with a bite boardor head strap.

Specific implementations. There are two basic typesof corneal reflex methods, depending on the location ofthe light source. In the first, and oldest, the light sourceis fixed with respect to the subject's head. To relateeye position to the material being fixated thereforerequires either a fixed head system, a method for record­ing head 'position (linear and angular), or a techniquefor recording the field of view relative to the head atevery sample. These methods are referred to as head­mounted or head-fixed corneal reflex techniques.

The second corneal reflex method fixes the lightsource in the target field rather than to the head. By

FIXATION

INTER·FIXATION MOVEMENT

RETURN SWEEP

REGRESSION

OURATION OF FIXATION

t ,, \\ ,\ \.·1 I

'. 1 /\ f\ I1 I\ \

t I) )

.:\ I\ ,, I\ I

l

404 YOUNG AND SHEENA

BEAM·SPLITIINGPRISM

Figure 6. Eye-marker camera tracks and records the eye'sglance. The image of a spot of light, reflected from the cornea,is transmitted by an optical system in the periscope through aseriesof prisms. This servesto superimpose the eye-markerimageon the scene image. The combined image can be monitoredthrough the viewfinder as it is photographed by the motion­picture camera (Thomas, 1968) (Copyright [1968] by Scien­tific American, Inc. All rights reserved).

output of the system is graphic, but a 15 by 15 resolvingdigital unit is available to extract the spot position froma television picture.

Head-mounted corneal reflex camera. An importantextension of the basic corneal reflex technique, in whichthe light source for the corneal reflex is still attached tothe head, is the head-mounted eye monitor camera. Inthis free head system, developed by Mackworth andMackworth (1958), a film or television camera that isaligned with the head during free head movementscontinuously records the field of view. The cornealreflex, measured as a reflected light spot from a periph­erallamp, as in the head-fixed system, is combined withthe visual field scene through a beam splitter and appearsin the film or video display as a white spot placed overthe portion of the scene which is being fixated. Thesystem is shown schematically in Figure 6. The accuracyis poorer than that with a fixed head system, approxi­mately ±2 deg, since it is subject to greater error fromthe relative movements of the light source with respectto the eye associated with head movements. Severalcommercial verions of the head-mounted Mackworthcamera are available. When the camera is mounteddirectly on the head, as in the early Mackworth helmet­mounted camera system, the interference with headmovements associated with the weight of the instrumentmay be considerable.

Recent advances have been aimed chiefly at reducingthe weight and moment of inertia of the portion of the

eye movement recorder which must be worn on thehead. Some improvement was made through miniaturecameras and miniature TV systems. The largest advance,however, has been the mounting of the camera separate­ly and carrying the visual information, including cornealreflex, and the field of view from the head to the TV ormotion picture camera via coherent fiber optics cables.An illustration of one such system is given in Figure 7in which the headpiece weight is reduced to 2 lbs. Notethat a bite board and headband are used for stabiliza­tion.

A similar fiber optics version of the Mackworth cam­era, manufactured by NAC, has a body which weighs lessthan l Ib, with another 5.3 oz for the fiber optics and9.90z for a camera adapter-not carried on the head.This version is pictured in Figure 8.

An alternative approach to fiber optics for droppingthe weight is the use of a miniature TV camera on thehead, combining the corneal reflex as before. A pictureof such a device and a diagram showing its principles ofoperation are shown in Figures 9 and 10. This system,manufactured by NAC of Japan, in conjunction with aTV system from Rees Instruments, Ltd. of England,requires a combining mirror in the field of view. Itsweight is 3lbs 130z with the camera shown, reduce­able to 150z on the head, including cable, with alighter camera according to specification sheets.

~~J

Figure 7. Mobile corneal reflex eye movement camera(Courtesy of Polymetric Company).

SURVEY OF EYEMOVEMENT RECORDING METHODS 405

Figure 8. Head-mounted corneal reflex illumination, viewing,and combining optics (Courtesy of Instrumentation MarketingCorporation).

Evaluation of corneal reflex methods. The uncorrect­ed linear range of all corneal reflex systems whichemploy a single light source for the reflex is limited toeye excursions of ±12-15 deg vertical or horizontal.Larger excursions place the reflex in the nonsphericaland rougher peripheral portion of the cornea and requirea complex (usually computer-generated) calibration andlinearization technique. The reflex range is ultimatelylimited by the size of the cornea and its partial disap­pearance behind the lids. In addition to head move­ments, other factors which limit the accuracy of cornealreflection methods to .5-1 deg are variations in corneashape and thickness of tear fluid, corneal astigmatism,

and the production of other reflections by eyeglasses(Hall. Note 4).

The output of most of these systems is usuallygraphic and therefore requires manual measurementand/or recording. But any of the systems whichproduce a single bright spot on film or directly on avideo signal can be used to provide conversion to thex-y coordinates of that spot. The most common ofthese converters detects the brightest (or darkest) spoton the video signal and determines its x and ycoordinates by the time of its occurrence. relative tothe horizontal and vertical scan sequences. Resolution

Figure 9. Head-mounted corneal reflex illumination, viewingand combining optics (courtesy of Instrumentation MarketingCorporation).

is about equal to two video lines, or about 1 part in250, for conventional TV scans. This information canthen be digitized for further processing of the part ofthe field being fixated. Polymetric Company makes alow-resolution digitizer OS by 15 matrix) for use withtheir instrument. No other commercial quantitativedevices exist for the simple corneal reflex method.

Limbus, Pupil, and Eyelid TrackingBasic techniques. The sharp boundary between the

iris and the sclera (the limbus) is an easily identifiableedge which can be detected optically and tracked by a

I,RELAYI OPTICS

,,-to-~~----- I COMPOSITE: OPTICS

SPOT LAMP ffi RELAYLENS FOCUS liJ OPTICS

t EYE I

~A~P--:--~ <@).~~ tflwPHOro LENS

'",.~-..::.::..' SEE THROUGHMIRROR

FiELDOFVIEW-------"

Figure 10. Schematic diagram for head-mounted TV cornealreflex system (Courtesy of Rees Instruments, Ltd.),

406 YOUNG AND SHEENA

Figure 11. TV picture of a bright pupil (Merchant &Morrissette, 1974).

variety of mea~~; a human observer looking at the eyecan. do. surprisingly well in determining where thesubject IS looking. If the entire iris were always visible~d not partially hidden by the lids, it would be asimple matter to trace its circumference and deter­~n~ its center. However, because only part of theInS ~s nor~ally visible, other optical methods, includingpupil tracking, are necessary to find its center.

When only horizontal eye movements are of concern,then the left and right extremes of the iris can betracked, either by measuring the gross difference inr~tlecte? illu~lination from fixed areas of the eye onelthe~ SIde ot .the ce~tral gaze position or by trackingthe limbus with a VIdeo scan system. When verticalmeasurement is also required, one may track eitherthe eyelid level, the pu pil position, or the vertical~lOtion of a ~isible part of the limbus. Nearly all~m~bus tracking systems use invisible, usuallyinfrared, illumination. They all measure the positionof the limbus relative to the photodetectors. Forhead-fixed photodetectors and illuminators, free headmovement is possible and the measurement is of eyerelative to the head.

The pupil offers a nunfber of distinct advantagesover the limbus. First, it is smaller and thereforeunobscured by the eyelid for a much greater range ofeye motion. For large eye motion, it presents to theobserver or the observing instrument a greater portionof the rou nd or slightly elliptical shape. The center ofthe pupil virtually coincides with the foveal opticalaxis of the eye. There is a 5-6-deg deviation betweenthe two, but with most measurement techniques, this~an generally be.calibrated out. The edge of the pupilIS usually a crisper, sharper boundary than thatbetween the sclera and the iris. This makes for ahigher resolution measurement.

On the other hand, the pupil, when viewed undernormal illumination, appears black and thereforepresents a lower contrast with its surrounding iristhan the iris does with respect to the sclera. This~ak~s !t a more difficult problem to automaticallydiscriminate the pupil. If collimated illumination isused, however, the light is reflected from the interior

~f the ~ye and to an observer looking along the illumina­tion aXIS, the pupil appears bright. This effect is oftenseen as "pink eyes" in flash photographs where the flashlamp is close to the camera lens. This phenomenon isemployed in the Honeywell oculometer and othersimilar devices (Merchant & Morrissette 1974) (seeFigure 11). '

Another characteristic of the pupil which providesboth an advantage and a disadvantage is the fact thatits diameter varies as a result of both psychologicaland physiological influences. This makes measuringthe center of the pupil somewhat more difficult but itdoes provide, in many techniques, the pupil diameteras a collateral output of the measurement which maybe of interest to the experimenter in correlating withthe position of the eye at any point in time.

Specific implementations: Scanning methods. Onetechnique involves scanning of the eye with a normalTV camera, which has sufficient sensitivity in the nearinfrared region (700-900 millimicrons) to be effectivewith IR lighting. Horizontal sweeps which intersectthe iris and the pupil have a video signal as shownschematically in Figure 12. The horizontal position ofthe limbus or the pupil is determined by thehorizontal sweep of the line crossing the center of theiris with video contrast very high. The time from thestart of the sweep to the drop in intensity at the firstiris boundary is a measure of the eye horizontalposition relative to the camera. Since the total sweeptime is 62 microsec, resolution of 1/500 of the linelength (comparable to vertical resolution) requirescircuitry to give time resolution of .1 microsec, whichis now within the state of the art. The choice of whichline is to be measured can be done in two ways. Oneway is to measure the left iris boundary every sweep

VIDEOSiGNAL

--t

MEASURE OF HORIZONTAL EYE POSITION(ONE SCAN liNE ACROSS EYE)

Figure 12. Scanning method for tracking limbus. FromYoung .(970) ~C?pyright 1970, McGraw-HilI Book Company.Used WIth permission of McGraw-HilI Book Company.).

SURVEY OF EYE MOVEMENT RECORDING METHODS 407

Figure 13. Various scan methods applied to tracking thelimbus and pupil.

PUPIL INTERSECTIONMETHOD

IMAGE DISSECTOR PUPIL

CIRCUMFERENCE TRACKINGMETHOD(Dissector 'urns righl inside pupil

and rett outside)

CENTROID OF VISIBLEPORTION OF IRIS METHOD

TANGENT LINE METHOD

returned reflected light at every instant. Cornsweet(1958) used such a flying spot scanner to track thelimbus. Rashbass (1960) used it to maintain a smallsweep tangent to the limbus at a fixed angularorientation. thus measuring the vertical andhorizontal eye position from tracking a single part ofthe limbus (see Figure 13).

A TV camera with an image dissector has beenapplied to precise tracking of the corneal ret1ex(Ishikawa. Yamakazi, Inabi, & Naito, 1970. Themain horizontal sweep is a normal full-pictureleft-to-right scan. which is used for approximatelocation of the bright reflex, whereas every secondsweep is a short (high resolution) line centered in thepart of the field containing the retlex. Thiscombination scan of coarse slow and fine rapid sweepswas used to track vertical and horizontal eyemovements over the range of ± 15 deg with aresolution of .01 deg and frequency up to IS KHz.

Specific implementations: Differential reflectionmethods. Scanning methods either require head fixationor the attachment of a scanning mechanism to thehead, directly or using fiber optics. Similar resultscan be obtained by measuring limbus position with twoor more small photocells viewing appropriate parts ofthe eye, either directly or examining its image. Just aswith the scanning systems, there are basically twocholces -either broad illumination and tightlyconstrained fields for photodetection or the use of

and select the one which is furthest to the left as themidline. The other method is to rely upon the verticaleye position measurement to select that horizontalsweep which crosses the center. For example,measuring the top and bottom sweeps which intersectthe pupil yields pupil diameter as well as vertical pupilposition. That line which is midway between the pupiltop and bottom can be selected as the midline and itsintersection with the pupil or iris measured (Sheena,1973) (see Figure 30). Vertical resolution is limited byscan line density. For standard US systems. thisyields, at best, I part in 212 per field (twofields/frame) at a 30-Hz rate. or I part in 525 at a60-Hz rate. Horizontal measurement of the pupilboundary has to take into account the varying pupildiameter.

Other methods of scanning have been applied totracking the limbus or the pupil. A programmablevideo scan, or image dissector, can be used to tracethe circumference of the pupil for determination ofthe center (Merchant, Note 5) or to trace only thevisible portion of the curve of the limbus, and fromthat information calculate the coordinates of itscentroid (Sheena, 1969). This latter method permitslarge eye excursions. Instead of illuminating the eyeuniformly and tracking with a wide scan, one can scana small spot of light across the eye and measure the

Figure 14. Differential reflection reading eye movementmeasurement device (Courtesy of Biometrics Division, NarcoBio-Systems, Inc.).

408 YOUNG AND SHEENA

-, CORNEA

- __~,~PHOTOCELL.~~ ,R2

( a )

IRIS

\(fj£>'\ T

ILLUMINATED AREASOF THE EYE

( b )

Figure IS. Light and photocell arrangement for limbus tracking (Stark, Vossius, & Young, 1962).

focused slits or circles of light on the eye with relativelylarge detection regions. Each of these methods is used,although the most common method uses broad fieldillumination so that only one source, and no mask, isrequired.

In the earliest versions of this device, Torok,Guillemin, and Barnothy (1951) and Smith andWarter (1960) imaged one side of the eye on a smallhorizontal slit placed in front of a photomultiplier. Asthe eye moved horizontally, the image on the slitcontained more or less bright sclera, depending on thedirection of eye movements, and thus the photocelloutput varied. Accuracies of 15 arc min over severaldegrees were obtained with careful head restraint. Acurrent commercial version of this technique is shownin Figure 14. This tracks the horizontal movement ofeach eye by recording differences in output of twophotocells on opposite sides of the iris image. AU ofthese methods require a fixed head and a fixed placefor viewed material. Richter (1956) placed the lightand photocells on goggles worn by the subject, and

Figure 16. Spectacle mounted differentialreflectivity device (Courtesy of BiometricsDivision, Narco Bio-Systems, Inc.),

relied on diffuse ret1ected light from the sclera ratherthan an image. Stark and Sandberg (1961) used twosmall lamps illuminating small discs on the two sidesof the iris and recorded the difference in reflectedlight from two photocells viewing these areas (seeFigure 15). The two cells permit a larger range of hori­zontal eye movement (to ±IS deg), a more linear rela­tionship than a single sensor, and relative insensitivityto vertical movements because of the push-pull sensorarrangement. With photodiodes attached to a spectacleframe, accuracy is over the order of 15-30 arc min. Whenthe head is rigidly held in place with a bite board, andnarrow beam photodiodes used with a range of a fewdegrees, resolution of 10 arc sec is possible (Zuber,1965). The interference of any changes in ambient illum­ination is now overcome in most systems by illuminatingthe eye with chopped light and then demodulating at thesame frequency. Chopper wheels in the light path areuseful in immobile systems (Wheeless, Boynton, &Cohen, 1966), but infrared light-emitting diodes(GaAs) are small and preferable for monitors that are

PHOTO DIODESPHOTODIOOES

IR SOURCENote. Arrows indicate modes of adjustment.

SURVEY OF EYEMOVEMENT RECORDING METIIODS 409

Figure 17. Limbus tracking with bifurcated fiber optics(Findlay, 1974).

built on spectacle frames to allow free head movement(young, 1970), as seen in Figure 16. Findlay (1974)used a bifurcated fiber optic bundle to both illuminatethe limbus and to collect the reflected light as shownin Figure 17. In this manner, both light source anddetector are effectively placed closer to the everesulting in a smaller illuminated area sensed by thedetector.

Measurement of the vertical position with thedifferential reflection methods is difficult, for thesame reasons as with scanning techniques. For thispurpose, Young tracked the upper lid position withthe output of two photodiodes arranged to be

PO

PO - Photodiode

LS - Lillht source

Figure 18. Tracking of the lower lid (Mitrani et aL, 1972).

IlL'ltT

Figure 19. Tracking of limbus image (Wheeless et aI., 1966).

relatively insensitive to horizontal motion. A newvariation on this method by Mitrani, Yakimoff, andMateeff (1972) uses a third photocell and a separatelight source to measure the vertical eye movements bytracking the related lower lid height. As seen inFigure 18, the light from this source falls directly onthe additional photocell, partially shadowed by thelower lid. Although linearity is a problem for large eyemovements, resolution of 15 arc min, comparable tothe horizontal, is reported.

Finally, the differential reflection technique can beextended to measurement of vertical eye movement bythe proper positioning of the viewing areas or the useof additional areas. Wheeless et al. (1966), in animmobile system with the head fixed using a biteboard. brought an image of the eye onto two detectorswhich could be arranged for horizontal or verticalrecording. The eye illumination, coming from achopped source, was brought to the eye with four fiberoptic bundles and their projection lenses, as shown inFigure 19. The eye viewed through a beam splitterwith visible illumination is imaged on a frosted screen.The four tiber bundles are then manipulated to yieldthe pattern of four slits on the eye seen in Figure 20,and the infrared filter is replaced for recording. Thedifference between the Iris-Pupil Detectors 3 and 4 is.a measure of vertical eye position, independent of

Figure 20. Position of detector fiber bundles on eye imaJll(Wheeless et aL, 1966). Area I and 2 are used for horizontal­motion detection, 3 and 4 for vertical-motion detection.

410 YOUNG AND SHEENA

limbus-Q photo- receptive

~'* area• : 0. •

, .lamp

photocell

Figure 21. Illumination pattern for two-dimensional limbustracking (Jones, 1973).

pupil diameter to first order. Another variationdescribed by Jones (1973) uses only two photocells.Using very short focal length lenses (12 mm) andrectangular masks in front of each photocell, heprojects 3.8 x .5 mm strips of photocell fields onto thelower limbus in the orientation shown in Figure 21.Horizontal eye motion effects the light sensed by thetwo photocells differentially, while vertical motionchanges them similarly. By separately adding andsubtracting the two photocell outputs, a monotonic,though nonlinear measure of vertical and horizontaleye position is achieved with a simple light spectaclemounted device.

Evaluation of limbus and pupil trackingtechniques. Many implementations of thesetechniques yield good results with good accuracies fora reasonable range. The output is an electrical recordwhich can have a good frequency response.

As noted, however, vertical eye movements are aproblem. The technique also requires head fixing or ahead-mounted device. The latter can be quite lightbut precludes eyeglasses. Iris coloration can also be afactor in its utilization.

Contact Lens MethodBasic techniques. The most precise measurements

of eye movement are made with one of the techniquesemploying some device tightly attached to the eye witha contact lens (Ditchburn & Ginsborg, 1953; Riggs,Ratliff, Comsweet, & Comsweet, 1953; Yarbus,1967). Conventional corneal lenses are too mobile tobe of use, and all measurement systems use speciallenses consisting of two individually ground sphericalsurfaces to fit snugly over the cornea and sclera. Foraccurate recording, it is essential for the lens to movewith the eye-both in steady displacement and duringthe high acceleration associated with voluntarysaccades. Tight fit and lack of slip is achieved by closegrinding tolerances and by the suction effect of20 mm Hg, or more negative pressure, between the

contact lens and the eye. Fender (1964) pointed outone way of developing the negative pressure by fillingthe cavity with a 2 % sodium bicarbonate solutionwhich osmoses out through the tissue. Yarbus (1967)achieved the accurate tracking required for imagestabilization work by withdrawing a small amount oftluid through a valve after applying the lens.Withdrawing air from under the contact lens is alsoeffective for stabilization but of limited time usebecause of the lack of corneal irrigation. All of thecontact lens systems cause discomfort. and the verytight ones usally required the application of a topicalanesthetic. Fender (1964) estimates that even with thebest current stabilization, the lens slips behind the eyeabout I arc min during a l-deg saccade and6 arc min during a 9-deg saccade. The contact lensand its associated attached material should not be of asize or mass to interfere with normal eye movements.Those which include a protruding stalk precludenormal blinking.

Specific implementations. The most commonlyused contact lens system is the "optical lever," inwhich one or more plane mirror surfaces ground onthe lens reflect light from a light source to aphotographic plate or photocell or quadrant detectorarray. (The latter is a solid state photodetectorarrangement which produces a pair of voltagesproportional to the x and y coordinates of a spot oflight falling on the sensitive area. It is used in thedouble Purkinje image method discussed below.) Thereflection from a plane mirror has a number ofaccuracy advantages over the corneal reflectiontechniques. First. and least important of all, for theplane coruu 't lens mirror, the change in retlectionangle is twice the eye rotation angle rather than 1.3times as large. Second, the imperfections related tothe cornea are eliminated. Most important. however,is the fact that the angle of reflection depends only oneye rotation and is independent of pure lineardisplacement. as long as the incident beam stillilluminates the mirror. This fact makes the systemlargely unaffected by the inevitable small headmovements which interfere with corneal reflectionmethods. Collimated light is reflected from such aplane mirror and focused with a lens on an imageplane. When the eye moves, lateral motion of a planemirror in a collimated beam does not cause a shift inimage position. Only rotation of the eye and themirror will produce deflections of the projected image(Ratliff & Riggs, 1950). Nevertheless, because of thehigh inherent accuracy of the system, very carefulhead stabilization relative to the recording device isusually used.

One early version of the optical lever principle, byDitchburn and Ginsborg (1953), used a 3-mm-diamoptical plate on the lens, 35 deg off the visual axis, toretlect a slit of light through a cylindrical lens camerato moving film for a record. The orientation of the slit

SURVEY OF EYEMOVEMENT RECORDING METHODS 411

Figure 22. Plane minor attached to scleral lens (Boyce &West, 1968). Haptic contact lens with stalk/minor unit attached.0-0' optical axis; R to suction reservoir; T polyethylene tube1/16 in. o.d.: M flat front surface minor; P grease pad (M maybe moved laterally on P for collimation purposes); S stalk 1/8 in.o.d., drilled bore 1/16 in. i.d., to allow entry of suction tube T;C corneal section of lens; L transcurves over limbus; H hapticsection; D Durofix seals;U cup.

and of the camera determined whether horizontal orvertical eye movements were recorded. Movements assmaIl as 5 arc sec can be recorded with the opticallever, over a range of ±S deg. The frequency responseof the system is generally determined by the stabilityof the lens in foIlowing the eye and by the recorderresponse. When the plane of the mirror is not normalto the visual axis, torsional movements of the eye alsoeffect the retlected beam angle. Matin and Pearce(1964) used this fact to advantage. He embedded twomirrors into the contact lens, one temporal and onenasal, so that the reflected beams move in oppositedirections during ocular torsion. By resolving thebeam detlections from an eye movement intoorthogonal rotations abou t the original axes with ananalog computer, they were able to measure three axisrotations to within 2 arc sec.

Those techniques using flush embedded mirrorssuffer from changes in mirror properties with tearfilm, especially when the output is measuredphotoelectrically rather than photographically. Forthat reason, several contact lens optical mirrorsystems employ mirrors or lights mounted on stalksprojecting from the lens. Fender (1964), in oneversion, attached a single mirror to the end of a stalkfor two-axis recording and at other times usedorthogonal mirrors for three-axis measurement.Construction details of one haptic contact lens unitwith a stalk/mirror are shown in Figure 22. A smalllight, weighing less than SO mg has also beenmou~ted at t~e end of a contact lens stalk, parallel tothe VIsual aXIS, to throw a moving shadow across a

photom uItiplier face for accurate recording ofhorizontal or vertical miniature eye movements(Byford, 1962). Byford showed that his aluminumstalk and lamp assembly had mechanical resonancepeaks at greater than 200 Hz (the one chosen was at350 Hz) and were adequate for the calculated ISO-Hzbandwidth required for miniature saccade displacementrecording. Further tests using a 1/8-in. o.d. hollowPerspec stalk showed it to have a high enoughresonance peak for satisfactory use for retinalimage stabilization. One way of creating a small lightsource on the contact lens without a lamp and itsassociated power leads, and thereby permitting lidclosure, is by using a small radiating source. Nayrac,Milbled, Parquet, Leclerq, & Dhedin (1969) placed asmall bit of the luminescent phosphor containingradioactive tritium on the contact lens. The l-rnrn!surface glowed brightly under ultr~violet illumina­tion. The position of the spot, and thus the eye, wasmonitored in two axes by use of two opticaldensity wedges, placed in front of separatephotomultipliers for recording horizontal and verticalmotions. They report a resolution of 20 arc min,which is poor for a contact lens system, but a usefulrange of .3 to 30 deg.

A recent addition to the repertoire of opticalcontact lens methods is one in which a multiple linepattern. fixed to the contact lens stalk, is imagedthrough a slit on moving film, as shown in Figure 23.The film pattern of one thick and several thin lines,shown schematically in Figure 24, contains all theinformation needed for measurement of vertical,horizontal. and torsional eye movements as a functionof time.

The princi pal non optical contact lens measuringmethod is the search coil technique introduced byRobinson (1963). Two small wire coils, orientedperpendicularly to each other and embedded in thecontact lens, pick up an induced voltage from twolarge perpendicular electromagnetic coils surrounding

1'Igure 23. Line diagram fixed to the eyeball by means of thescleral contact lens (Forgacs, Jarfas, & Kun, 1973).

412 YOUNG AND SHEENA

Figure 24. The modified photonystagmographic method ofForgacs (Forgacs et at., 1973). When properly set, on the surfaceof the film cassette facing the lens (4) of a special camera(3) provided with .1-mm-wide transversal slit (6) the magnifiedshup picture (5) of the line diagram fixed to the eyeball (1) bymeans of the scleral contact lens (2) is obtained. This picture iscrossed by the transversal slit (6). The thick line of the linedrawing is perpendicular to the slit, while the thin lines intersectit at an angle of 30°. On the film moving downwards (8) in thecassette behind the transversal slit at a constant speed of10 mm/sec, curves in a number corresponding to the number ofpoints of intersection between the lines of the diagram and thetransversal slit are depicted as the eye moves. This set of curvesmakes up the photonystagmogram (7).

the subject. The induced voltage in each coil variesonly with the sine of the eye angle relative to themagnetic tield and is independent of head positionwithin the uniform portion of the tield. Use ofdifferent frequencies or quadrature phases in the twolarge coils permits the detection of the angle betweeneach search coil and each driving plane and,therefore. stable three-dimensional angular localiza­tion to the order of I arc min over large angularexcursions. Voltages are read out through lightflexible wires extending from the lens. Extension ofthis technique to implanting the coils on the globe forchronic animal experiments permit the preciserepeatable localization of eye position.

Evaluation of contact lens methods. Although thecontact lens systems offer the best resolution of anysystem down to 5-10 arc sec, they do so in general atthe sacrifice of range. They are normally applicablefor the study of miniature eye movements and, withthe exception of the magnetic search coil method. areinappropriate for movements of greater than 5 deg.The expense and discomfort of the contact lens makes

it a technique more suitable for usc on a few subjectsfor physiological studies of fixation than for wide­spread investigation on the normal population.

The dangers of litting a contact lens with negativepressure arc also considerable. There is the possibilityof deforming the cornea and the worse hazard ofdamaging the accommodation muscles as a result ofthe pressure stress placed upon them.

Point of Regard Measurement: Tracking of CornealReflection Center with Respect to the Pupil Center

General principle. It is often of interest to theinvestigator to know the fixation point of the subjectas it falls in space rather than the position of the eyewith respect to the head. In other words, one wants todetermine where a subject is looking regardless ofwhether his eye got there through eye rotation or headmot ion. Clearly. the two measu rements are equivalentif the head is fixed; or the position of the head may bemeasured by any of a number of techniques and theposition of the eye in space deduced by summing therelative position of the eye to the position of the head.Various head-position measurement techniques arcdiscussed in a later section of this paper.

One of the main problems of measuring eyedirection optically is that of separating lateral motionof the eye relative to the observer or the sensor androtary motion of the eye relative to the scene. Forexample. eye rotation can be measured by trackingthe corneal reflection of a suitable source; however. itis not possible to distinguish this from motion of thecorneal reflection which results from translation. Unlessthe corneal reflection tracker is the head-mountedvariety described in a previous section, lateral motionof the head will introduce large errors-approximatelyI deg of error for every 5 x 10-3 in. of lateral headmotion. It is difficult to eliminate this head motioncompletely by any fixing methods.

A desirable method is one which would allowrelatively free. natural head motion and measure theposition of some parameters of the eye in such amanner that they would indicate eye rotation only andconsequently point of gaze in space. This may be doneby measuring features ofthe cye that onlv change withrotation, or by measuring the positions of various detailson the eye which move differently as a function of headmotion and eye rotation, such that pure eye rotationcan be deduced from them. Two features which satisfythis requirement are the corneal reflection and thecenter of the pupil. Each has been independently usedfor eye-position determination, but, differentially,they provide a powerful tool.

Specific implementations.The corneal retlexwas discussed at length in a previous section. Itpossesses a very useful property. If a su bject is look ingdirectly at a light source, and an observer is looking atthe subject's eye from a position at or very ncar the

SURVEY OF EYE MOVEMENT RECORDING METHODS 413

Figure 25. Photograph from wide angleMackworth camera (Courtesy PolymetricCo.).

..t~.".. "~.~~~1t\t ..~'~'4.'(\

, ..::•..•...,.\... '"1· i ·•.; ...• /'..,. ... '.

Eye of two-year-old child,Fixationonsmall square.Four-sided grid marker used.Stimulus material isrear-illuminated transparency.

light source, then, from the obvious considerations ofsymmetry, the corneal reflex will appear to theobserver to be in the center ofthe subject's pupil. Thisresult may be used in either of two ways. First. theimage of the point on which the subject is fixating willbe found in the center of the pupil. Second, thesubject's angle of gaze with respect to the light sourceis approximately proportional to the distance betweenthe image of the light source and the center of his pupil.These two properties are equivalent to each other. In thefirst, multiple light sources are used, and point of regardis determined by which light source is imaged in thecenter of the pupil. In the second, a single light source isused, and the error between its image and the center ofthe pupil is measured. Both results are monotonicallyrelated to the point of regard. Measurement of thecenter of the corneal reflection from the limbus is alsopossible but poses practical problems.

Both these approaches are applied to methods ofeye point-of-regard measurement. The first iseffectively the inverse of the conventional cornealreflex method where the reflections were superim­posed on the scene being viewed by the subject, andthe eye position is determined by where the cornealreflection falls on the scene. In the wide angleMackworth camera, the illuminated scene is itselfused as the light source for the reflection. Byphotographing the eye at high magnification, thecorneal curvature produces a superimposed image ofthe field being viewed, with the position "being"fixated lying directly in the center of the pupil (SeeFigure 25). Since the entire field is being imaged overthe small area of the cornea, resolution is lost relativeto the simple corneal reflex method. This method istherefore not very amenable to high resolutionanalysis. Also, in order to photograph a scenereflected from the cornea, the field must containbright light sources against dark backgrounds. Itdoes, however, provide a method of obtaining eyefixation points without excessive head restraintrequirements. All that is required is that the pupilcenter be visible to the observer or the observinginstrument.

A commercial instrument which performs this type

of measurement is the Polymetric Company wideangle eye movement recorder. This instrumentoutputs a photograph (Figure 25) from which fixationmay be determined, or a 4 x 4 digital position outputprovided with a television system. It offers a 40-degmaximum field of view, ±2.5 recording accuracy,when manual determination is made, and a I part in 4electronic accuracy. The sampling rate is 12photographic frames/second, or 60 TV fields/second.The system requires no bite board. The subject looksat the scene through a viewing aperture.

In another application of the wide angle cornealreflex technique using multiple light sources,Salapatekand Kessen (1966), and later, Haith (1969),developed a wide angle corneal reflex technique foruse with the human infant. The arrangement of fixedinfrared illuminators over the infant's head is shownin Figure 26. The TV picture of the eye contains thecorneal reflex of some or all of the six infraredilluminators; scoring is done using video tape on aframe-by-frame basis.

The x-y coordinates of the pupil center, and of aparticular infrared reflection, are measured with a

TELEVISION CAMERA

BAUSCH AND LOMP

I:"'LUMIN:.":"ORS

""INFRARED fiLTERS

....... STIMUlUS SCREEN

Figure 26. Illustration of a system for recording eye behaviorin infants ([Haith, 1969) Copyright (1969) by the AmericanPsychologicalAssociation. Reprinted by Permission.).

414 YOUNGAND SHEENA

(d)

(b)

KSin e[ ~:~;::=e~~Pupil andCorneal Pe f Lec t ion

(c)

(a)

Figure 29. Effects of eye translation and rotation on cornealreflection-pupil center (Merchant &: Morrissette, 1974): (a) Eyelooking straight ahead-note corneal reflection is at center ofpupil. (b) Eye looking straight ahead but laterally displaced­note corneal reflection still at center of pupil. (c) Eye lookingto side-corneal reflection displaced horizontally from pupilcenter. (d) Eye looking up-corneal reflection displacedvertically from pupil center.

Radiation from Source WithinOculometer Reflected at Cornea

Figure 28. Displacement of corneal reflection from center ofpupil, K sin 8, is proportional to the angular direction, 8, of theeye and is independent of the position of the eye (Merchant &:MoRissette, 1974).

Morrissette, 1974). This phenomenon may begraphically seen from Figure 29, which shows that thecorneal reflection with respect to the center of pupilposition remains unchanged as a result of lateral headmotion and changes for eye rotation only.

A number of versions of this instrument exist whichautomatically find the pupil center and' cornealhighlight in the x-y plane and calculate eye positionfrom the relative vector. These include the Honeywell

Centercenter of ofCorneal Pupil

Cu~ture !- ~ -'jk-----f--c:-----""T"'----

I.... (,.,

_ <,~(~~~S~i~~~~~Virtual_

~~:"rc.J

lri

v:O(O rAPE IIt£COROEAT,v, lilOfHTOP:

,rw'\'..~ I\ ...--/••• • •

An illustration of a procedure for recordingeye-to-eye contact between mother and child. (A mirrorat a 45-degree angle to the visual axes of the mother andthe baby reflects visible light thus producing an apparentnatural vis-a-vis image to both. The infrared image ofthe infant's eye is transmitted through Mirror Y, reof1ected by Mirror X, and then recorded by the top tele­vision eye camera. The infrared image of the mother'sface is transmitted by Mirror Y and recorded by thelower television' face camera. The outputs from thesetwo cameras are mixed into one picture and then recordedonto video tape.)

Figure 27. An illustration of a procedure for recording eye­to-eye contact between mother and child (Haith, 1969) (Copy­right (1969) by the American Psychological Association. Re­printed by permission).

cursor whose position is outlined by a superimposedcross on the video monitor. By use of a video mixingsystem, a second TV picture can be added to the firstso that each frame contains both eye position and apicture of the field being scanned, as shown inFigure 27.

The wide angle Mackworth camera uses whateffectively is a wide field corneal reflex and determineswhich part of it appears in the center of the pupil. Asecond version of this method is to use a single pointcorneal reflex and to find how far it lies from thecenter of the pupil. This principle was put intopractice by the oculometer, developed by Merchant(Note 5) at Honeywell. In the oculometer, a singlelight source is used to produce a corneal reflex on thepupil. The problem of obtaining net eye motion withrespect to the scene is overcome through the trackingof two elements of the eye that move differently witheye position and head position. As discussed above,the pupil is viewed by observing optics that are coaxialwith the illumination optics in the oculometer. Thecorneal reflection appears always in line with thecenter of corneal curvature. The apparentdisplacement of corneal reflection from the center ofthe pupil is thus equivalent to the apparentdisplacement of the center of corneal curvature fromthe center of the pupil, which is obviously a functionof eye rotation only. See Figure 28 (Merchant &

SURVEY OF EYE MOVEMENT RECORDING METHODS 415

Figure 30. Operator indicators super­imposed on TV image of eye as a functionof threshold setting showing status ofmeasurement. Measurement is good whenpupil is properly delimited (CourtesyWhittaker Corporation).

oculometer (Merchant & Morrissette, 1974), aportable oculometer for transportation studiesdeveloped at the Department of Transportation(Davis, Note 6), the EG&G/Human EngineeringLaboratory facility (Monty, 1975; Hall, Rosenberg, &Monty, Note 7); a laboratory oculometer developed bythe University of Alberta (Petruk, Note 8); and theWhittaker Corporation eye view monitor (Sheena,1973).

Several techniques are available to present anonvisible light source to develop the cornealreflection and to locate the pupil center and cornealreflex on the image. Honeywell and Whittaker use aninvisible infrared light source for the cornealreflection and use a TV camera with sufficientsensitivity in the infrared region to detect thishighlight easily. EG&G. in their original version, usedvisible light. By having all of the light source in the roompolarized, except for a small portion and by passingall returned light through another polarizing filter to alow light level TV camera. the corneal highlight wassharply defined.

With the Honeywell oculometer, and others. thepupil is backlighted; and the resulting image is thebright pupil and even brighter corneal reflex as inFigure II. In some Whittaker units. and theEG&G /Human Engineering Laboratories system, thepupil is black as in Figure 30.

The center of the pupil was determined inMerchant's original design by an image dissectorwhich traced the circumference of the pupil andtracked it as it moved. As with any image dissectortracking system, this required some time forreacquisition of the pupil whenever it was lost. as. forexample, during a blink. However, currently, all ofthe systems appear to use a conventional TV rastersystem in which the geometrical coordinates of thepupil center and of the highlight are determined bytiming signals on the video scan. For example, if thecorneal reflex corresponds to the brightest point in thepicture, and it was located in the upper left-handportion of the picture a quarter of the way down fromthe top and a quarter of the way in from the left, thehighest voltage on the video scan would occur 25% ofthe time from the beginning of a new field followingthe vertical sync and 25% of the time across ahorizontal scan line. A variety of processing andpattern recognition aids in determining the center ofthe pupil and corneal reflection correctly have beendeveloped in connection with video systems.

Indicators are also superimposed on the video asoperator aids; see Figure 30 (Sheena, Note 9). Asilicon vidicon television system is also generally usedfor its high sensitivity, especially in the IR region.

The most advanced Honeywell oculometer canfollow large head motion by maintaining a field ofview which covers the subject's eye by driving atwo-axis mirror which compensates for gross headmotion. The position of the head when the eye is lost isdetected through a search pattern (See Figure 31).

With the EG&G/HEL system, the subject sits in achair and views a projection screen without evenknowing that his eye is being observed with a cameralocated underneath the screen. This camera leads to avideo processing system also attached to a PDP-IIcomputer. Eye movement and fixation information isextracted and recorded in any of a number of formats,including video tape recording and others (SeeFigure 32).

A variant or the basic single corneal reflection-pupilcenter vector method is used by Hochberg (Note 10).He employs two corneal reflection spots to eliminatethe effect of distance variation from the optics to thesubject or overall optical system magnification. Theseparation between the two spots becomes the basiclength around which all the other measurements arenormalized. Th is therefore eliminates the need forabsolute calibration, The pupil center position is thenmeasured with respect to a point equidistant betweenthe two corneal retlection spots.

Most of these systems use a digital computer.Sophisticated processing is therefore possible andcorrection and linearization for various limitations ofthis measurement principle can be overcome. Inparticular, the corneal retlection method is generallylimited by the curvature of the corneal bulge to less± I5 deg. Beyond that. the cornea flattens out and themeasurement becomes nonlinear, although stillmonotonic. There may also be imperfections andasymmetries in any individual cornea that makes themeasurement nonlinear; changes in the tear film alsocause a problem. As the pupil constricts and dilates, itdoes not normally maintain a fixed center with respectto the eyeball. All of these problems can be eliminatedto a large extent by the processing which isincorporated in the computer and circuitry of thesystems.

Any system which determines the pupil center foreye movement measurement also can make pupil areaor pupil diameter available as an additional output

416 YOUNG AND SHEENA

I

ILLUMINATION OPTICS

LJ

AV

TUNCST9..._HP__w -t>I-__-'~

SILICONVIDICON

TV CAMERA

SENSOR UNIT

MIRROR *CONTROL

EY[·SP\CE1'«)T10N' 80\

SIGNAL PROCESSOR(HINICOHPlrrER

WITH YIDEC>-OIGITAL!---o-lINTERFACE)

A-D AND

CON~~:TER *HIRROR AND

FOCUS

DRIVE *ELECTRONICS

I IANALOG OtTrPUT SIGNALS

t tPUPIL DIAI£TER*These items deleted in cubic inch version

--------EYE DIRECTION

Figure 31. Schematic of the cubic foot remote oculometer (Merchant & Morrissette, 1974).

Figure 32. EG&G-HELoculometer system (Mont), 1975).

SURVEY OF EYE MOVEMENT RECORDING METIIODS 417

Figure 33. Simple field representation of cornea and rearsurface of lens (Cornsweet & Crane, Note 11). Location of thefirst and fourth Purkinje images for (a) collimated light on theeye axis and (b) collimated light at angle .a. from optic axisofthe eye; EA, eye axis; FT, fust Purkinje image; FR, fourthPurkinje image; L, lens; C, cornea. The dark section of arcis the equivalentmirror for the fourth Purkinjereflection.

surface of the lens, and a fourth occurs at the rear sur­face of the lens where it interfaces with the vitreoushumor. The second Purkinje image is relatively dim, thethird Purkinje image is formed in a plane far from theothers, and these two are not used in this measure­ment method.

Cornsweet and Crane (1973) used the first and fourthPurkinje reflections as the two features of the eye whichthey track. Like the relationship between the cornealreflection and the pupil center, these two marks movetogether under eye translation but differentially undereye rotation.

For purposes of simplicity, the two surfaces in ques­tion-the front of the cornea and the rear of thelens-are assumed to resemble a clam shell arrangement(shown in Figure 33) where both surfaces have the sameradius of curvature and have a separation equal to theirradius of curvature. C\ is the center of curvature for thecornea, and C4 is the center of curvature of the rear ofthe lens. If these two surfaces are assumed to be spheri­cal, the effect of collimated incident light is to produce

with virtually no additional electronics processing.With continued interest in pupil diameter as ameasurement of tension or arousal, this techniqueoffers the ability to monitor the effect on pupildiameter of different portions of the scene beingscanned.

At the present time, there are two known systemscommercially available: (1) The Honeywell Oculo­meter Mark II, which allows a cubic inch of motion,and the Mark III, which allows a cubic foot ofmotion; (2) the Whittaker Corporation eye movementmonitor is also commercially available and allowsup to a foot of head motion. The Honeywell unitprovides an accuracy of 1 deg, for a maximum rangeof 60 deg horizontal and 40 deg vertical, with timeconstant of .1 sec. The Whittaker unit provides anaccuracy of 1 deg, for a range of 40 deg horizontaland 30 deg vertical, with a sampling rate of 60/sec.Both systems also provide instantaneous pupildiameter.

Evaluation of the corneal reflection-pupil centermeasurement. There is no question that maintenanceof a fixed head or attachments to the head aredifficult, uncomfortable, unsuitable for many subjects,and may require long setup time. Therefore,the methods that measure eye point of gazerelative to space without requiring any head positionmeasurement or stabilization, offer considerableadvantage. They are comfortable; the data isgenerally in a form very amenable for processing; andthe illumination is usually infrared and, therefore, notannoying or disturbing. Many of the units alsoincorporate a great deal of information and signalprocessing, which allows broadening the limitation ofeye movement measurement to a greater range of eyemotion and even larger head movement allowances, withsignal improvements by averaging, linearization, andvarious other corrections.

This advanced and sophisticated instrumentation ismore expensive than the various other simpler, moredirect methods of eye movement measurement. It isalso generally bulkier and more involved to learn touse, although easier once learned.

What these methods may lack is speed; the televisionsystems can operate only as fast as 60samples/second. The basic speed limitation comesfrom the lag of the TV camera tube. Also, precisionsand accuracies are not as good as those that can beobtained with head-fixed contact lens, limbustracking, or corneal reflex methods.

Measurement of Eye Rotation by the Double PurkinjeImage Method

Principle. As light passes through the eye, reflectionsoccur at the various interface surfaces. At the surface ofthe cornea, there is the well-known corneal reflection orthe first Purkinje image, a second one occurs at the rearsurface of the cornea, a third one occurs at the front

CR

CR

EA

\FH

418 YOUNG AND SHEENA

Figure 34. Schematic layout of the double Purkinje imageeye tracker (Comsweet &: Crane; 1973). Schematic of theeye-tracker optical system: VT, visual target; R, allowed rangeof eye movements; lA, input axis; CA, collection axis; CA',extension of collecting axis; S, light source; S" artificial pupilimage at pupil of eye; CW, chopper wheel; S., source of Purkinjepattern, imaged at infmity; DC, dichroic mirror; M, front surfacemirror; Mx and My, motors that drive M in x and y direction,respectively; BS, beam splitter; P, and P4' quadrant photocells,A4 , aperture in front of P4' Focal lengths of lenses L.,La' and L4 are 60, 150, and 90 mm,

the two images roughly in the center, equidistantbetween the two surfaces. The distance that each imagemoves as a result of eye rotation is related to the dis­tance from the center of rotation CR to the center ofcurvature for the surface that forms it. The separationof the two images as a function of eye rotation ~ is ap­proximately given by S = 7 sin a, where S is in milli­meters.

Change in separation between these two images there­fore is directly related to the angular rotation of the eyeand independent of head translation.

Implementation. Figure 34 shows the optics employ­ed for this kind of measurement. From a light sourceS, Circular Aperture S2 forms two Purkinje images onthe eye. Collection Optics L4 view the eye and image

B

\

\

\

\

/\CA' \

\

\

o

s

the two Purkinje reflections on the two photodetectors,PI and P4. These are four-quadrant photodetectorswhich yield a signal proportional to the position of theimage off center. They control servomotors to drive eachimage continuously to center. The final output is thedifference in the electrical signals that are generatedin the servomechanisms required to maintain a centerednull condition. As the eye translates, both images movein the same manner, and the difference is effectivelyzero. This is performed in both axes. The two Purkinjeimages are separated by the optics so that they can betracked independently. .. _

This instrument has been developing through a num­ber of stages. The recent units have allowable headmotion of ±.5 em in all three axes. Automatic focusinghas been included to allow head movements along theoptic axis. The field of view allows a coverageof ±15 degvertically and horizontally. Although very precise align­ments are necessary to obtain the proper image position;the illumination optics have themselves been controlledto track the motion of the eye and place the illumina­tion on the pupil.

The bandwidth of the system, which is limited by theservomotors which drive the photodetectors, is quitehigh-up to 300 Hz. The instrument also boasts an accu­racy of 2 min of arc in response to I deg of step.

Evaluation of the double Purkinje image eye tracker.This technique represents a new and promising way ofmonitoring eye movements. It is the only head­movement-independent eye-tracking method which canmeasure eye position with such accuracies. It has ahigher frequency response than the television systems. Inthese ways it is superior to the pupil center-cornealreflection center method.

It is planned to integrate accommodation measure­ment using the Stanford Research Institute optometer(Cornsweet & Crane, Note 11), into this system so thatthe power to the eye is also measured along with eyerotation for what is effectively a three-dimensional eyeposition determination.

The eye tracker falls slightly short of other techni­ques in the field of view (±IS deg) which is pupil­diameter dependent. It also requires higher illuminationthan the other techniques in order to bring the fourthPurkinje image above the noise. The allowable headmovement is, still, not very large; but this may also befurther increased. It appears also that the optics need tobe fairly close to the eye. What the system cannotprovide is pupil diameter, but this may also be laterincorporated.

Normally for high-precision measurement using thisinstrument, the head is stabilized with a bite board.Some experiments usually requiring contact lenseshave been performed with this device.

SURVEY OF EYE MOVEMENT RECORDING METHODS 419

Figure 35. Apparent ellipticity of pupil as eye rotates.

Measurement of Fixation Point by Determining theRotation of a Plane Attached to the Eye

General principles. In this section are grouped themethods which measure some parameter of the eye thatvaries with eye rotation but is for the most part inde­pendent of head translation.

These methods include measurement of the orienta­tion but not translation of some imaginary or real"plane" attached to the front of the eye relative to afixed observer. One way is to determine the apparentellipticity of the pupil to an observer. Other ways areto attach just such a plane to the eye, a plane mirrorwith a contact lens or the plane of a search coil imbed­ded in a contact lens (Robinson, 1963).

Another method invariant to head motion is thetracking of an image focused on the retina. If an eye isfixating a particular point in space, then that point willalways fall on the fovea regardless of head translation oreye rotation. This principle was put into practice by aretinal image tracker developed by Cornsweet (1958).Again, the principle here is the measurement of rotationof a plane attached to the retina.

Specific implementations: Ellipticity of the pupil.When the pupil is viewed head on, it appears circular. Ifthe eye rotates, the pupil will have an apparent ellip­ticity. The apparent shape of the pupil to a fixed observ­er will not vary as the head of the subject moves butonly as the subject's eye rotates. If the eccentricity ofthe pupil is measured, therefore, it would indicate therotation of the eye only. One problem, however, isthat the measure is an even function of eye position,and some other course determinants would have to beused to differentiate between up anddown and rightand left. A second problem is that a rather involvedgeometrical computation would be required. SeeFigure 35. There have, therefore, been no successfulimplementations of this method (Hall, Note 4).

Specific implementations: Contact lens method.

A plane mirror or a search coil attached to the eyeprovides a geometrical "plane" whose orientationmay be determined independent of translation. Thesetwo approaches were discussed in detail in the sectionconcerning specific implementations of the contactlens methods.

Specific implementations: Tracking of a retinalimage. The retina is usually observed under wide fieldillumination, but if the field is focused to a small spot onthe retina and the subject is fixating it, then the image ofthe illumination spot clearly remains on the same pointon the retina. Clearly, also, the retinal image of a secondpoint near the first, and in the same plane, would befixed to the retinal image of the first.

Another way of viewing this concept is to consider asubject looking at a very distant object, say a star; theimage will not shift on his retina if he translates his head,but will obviously move if he rotates his eye. A colli­mated light source is equivalent to a very distant object.

Cornsweet (1958) had a subject fixate on one point,and he projected a small rectangle of light on the opticdisk as shown in Figure 36. If the relation between thefixation point and the spot is constant, the spot posi­tion on the disk shifts only with fixation changes. Apronounced blood vessel was selected on the disk, andthe relative shift of the imaged spot was measured by thedistance the spot had to be moved to be repositionedon the blood vessel. In other words, whenever fixationchanged, the spot went off the blood vessel and wasrepositioned there. This was actually accomplishedby a flying spot scanner which scans this vessel. Theinstantaneous reflected light was viewed with a photo­multiplier (See Figure 37). The time from the beginningof the scan to arrive to the reference and vessel wasdirectly related to shifts in fixation.

50 ~

..., f+ 200NIN ARC

Figure 36. Scanning of retinal vessel (Cornsweet, 1958).Drawing of the optic disk, right eye. The small rectangle is thescanning spot. The dashed lines indicate the locus of its path ofmovement.

420 YOUNG AND SHEENA

HIGH

LEFT

CENTER

OJ

tLnI·o 90 180 210 3608."

b,

Figure 38. Circular scan x-y tracker.

Current and advanced methods of monitoring headangle (head line of Sight, LOS), suitable for combinationwith eye measurement techniques to yield eye pointof regard, are given below.

All of the head measurement techniques, with theexception of the eye-tracking video method discussed inconjunction with the corneal reflex-pupil vector systems,require some attachment to the head. The attachmentsmay be as unobtrusive as reflecting surfaces or distinc­tive optical targets taped to the head, spectacles, or biteboard; or they may involve a helmet to carry the trans­ducers. Basically, five methods have been utilized:optical, ultrasonic, electromagnetic, inertial, and me­chanical. All but the last can be free of any mechanicalconnection between the subject's head and his surround.

Optical head position sensors. Optical x-y trackingtechniques that have been developed for tracking rocketsor small variations in thickness in industrial processescan be used to locate the x-y coordinates of a smallpoint attached to the head. The basic scanning methodsdiscussed under the corneal reflex implementation areapplied to locate the brightest or darkest spot in a sceneservo-drive the camera or mirrors to keep it centered.One common variation is the use of a circular scanpattern which uses phase-sensitive detection of the videosignal to determine the required direction to draw thecircle center for alignment over the target, as shownin Figure 38. These devices can be stabilized to resolveangular motions of a few seconds of arc with appropriateoptics. By using three such targets, all translations androtations of the head could be calculated.

The most common optical head tracking devicesemploy a helmet of some sort with an array of lightsources (or photodetectors) on it, and the cornpli-

IIlLT.I'OWIIlIU"","V

Figure 37. Fundus tracking system (Comsweet, 1958).

Head Movement Measurement to Obtain Point of RegardAll eye movement monitoring techniques fall broadly

into two categories-those that measure the position ofthe eye relative to the head and those that measure theorientation of the eye in space. For studies of the eyemovement control system, per se, the first categorytechniques normally suffice. When the concern is withidentifying the elements in a visual scene that are beingfixated or scanned, however, the orientation of the eyein space, or the "point of regard," is required. Any ofthe first category methods can be adapted for use in thefree head fixation pattern application by measurementof head angle and position (See Figure 43). Whether ornot head linear position is required in addition to angu­lar orientation depends, of course, on the freedom ofhead linear displacement permitted, the distance to thematerial being viewed, and the desired system accuracy.For this consideration, we also review some commontechniques for measurement of head position, all ofwhich can be combined with eye position recording toyield the point of regard.

The only eye movement methods which give theangle of gaze relative to a reference point in the visualfield are those with a camera mounted to the head, thosethat track two separate points in different planes on theeye, or those that measure retinal plane rotation. Any ofthe methods which track a feature on the front of theeye or EOG must be corrected for head position to givethe point of regard.

Resolution of 10 sec of arc was possible with thismethod. Resolution was limited primarily by instru­ment noise. Also, blood vessel diameter changes andcross-talk from vertical eye motion were problems.Large eye motion would move the spot off the diskand lose the measurement.

This is a very precise method not requiring a fixedhead. It seems to be limited only by the practicalproblems associated with viewing the retina. Technologi­cal advances in this area are certainly possible.

SURVEY OF EYE MOVEMENT RECORDING METHODS 421

(e)

Figure 39. Optical head position sensor (Chouet & Young,1974). (a) Helmet and disposition of eight photodetectors.(b) Helmet liner with two LEDs. (e) Monitoring equipment.

mentary devices fixed to the surround. One such system,built with the idea of measurement of astronaut headangles relative to a large visor or a fixed outer shell,is shown in Figure 39. It uses three miniature pulsedinfrared LEDs on the helmet liner and an array of eightphotodiodes as receivers (Chouet & Young, 1974). Thecross-coupling between head angle measurements invarious axes is minimized with a small computer andcompensation circuit.

The need for accurate knowledge of head position fora variety of "visually coupled systems" for hands-offlocalization and aiming has led to the developmentof accurate and reliable head line of sight sensing sys­tems. One such system, shown in Figure 40, was devel­oped for tracking the pilot's line of sight within the re­stricted confines of a cockpit (Ferrin, Note 12). Thesensor surveying unit (SSU), fixed to the laboratory orcockpit, emits two thin fan-shaped beams of IR whichsweep across the photodetectors in the helmet, trigger­ing each one at the moment of intersection. Use of theIR fans and detectors on each side of the helmetincreases the range of head motion allowed while stilltracking the line of sight. Although this and relatedsystems appear bulky because of the helmet and relateddisplay equipment, the actual hardware which must beworn on the head for head tracking is merely two orfour miniature photodetectors.

(b)(a)

y

Figure 40. Head lineoQf-sight sensing system (Ferrin, Note 12).

ROTATING FANSOFIR ENERGY

PHOTO DETECTORS

DETECTORNO. 1 {FRONTl

DETECTORNO.2 (REAR)

DISCRETE 1LIR BEAMPOSITIONS

/ ~ -- PARALLEL TO AIRCRAFT

~!.....3V~ LONGnUDINAL"""

z

(DOWN)FORWARD

-

422 YOUNG AND SHEENA

DETECTOIS

SOFTWAlfINTERfACE

LEDDIIVEIS

DETKTOIS

Figure 41. LED line-of-sight headposition measurement system (Haywood,Note 13).

COMPUTEIlEADOUT

INTEIFACE

1mA, ~, n DIIECTIONCOSINES

PITCH ANGLEYAW ANGLE

A newer design proposed for the helmet line ofsight measurement uses four LEDs mounted on a hel­met, each one pulsed at a different frequency(Haywood, Note 13)(see Figure 41). Up to four opticallinear position proportional detectors are positioned inthe laboratory fixed frame. Each detector measures thex-y coordinates of the image of each source in its fieldof view and thereby provides the computer with theinformation required for the resolution of the line ofsight angles of the helmet. The goal for such a systemis ±1 deg static accuracy in head pitch and yaw, overa range of ±75 deg about each axis.

Ultrasonic head position sensors. The use of mul­tiple ultrasonic transmitters and detectors on the helmetand in the laboratory permits head position to be sensedby measurement of the distance from each source toeach receiver by means of the sonic delay time. Prac-

Figure 42. Ultrasonic head positionmeasurement system (Sawamura, Note 14).

TRANSMITTERTRIAD

tical implementation of a multiple source system tomeasure head position and orientation was demonstratedat MIT's Lincoln Laboratory for computer graphics ap­plications. One possible arrangement of ultrasonic trans­mitters and receivers, shown in Figure 42, is especiallyuseful for pitch and lateral rotation head movementsensing (Sawamura, Note 14). The weight of the helmet­mounted portion can be reduced by mounting the trans­mitter, rather than the receivers, on the helmet. Testsof a similar system showed static errors of less than1 deg rms, with range of up to 360 deg possible. Theultrasonic sensing systems inherently provide head posi­tion as well as orientation information.

Mechanical head position sensors. Head position canbe measured by mechanical linkages to the target field.Perhaps the simplest is the attachment of flexible shafts,one for each axis of rotation to be measured, from a hel-

TYPICALTRANSMISSIONPATH, RECEIVER

TRIADI

SURVEY OF EYE MOVEMENTRECORDING METHODS 423

met to the laboratory bench. Head freedom of motion isrestricted somewhat by the cables, depending on theirnumber and length.

The next step in mechanical measurement is a formof goniometer for measuring angles of one body relativeto another. These typically consist of rigid or telescopingrods with potentiometers at each linkage to measurehead angles. Sperry developed such a rigid linkagesystem with a magnetic quick disconnect for pilothelmet orientation measurement. A two gimbal systemwith a vertical and a horizontal potentiometer wascombined with a limbus reflection eye movement mon­itor to yield a practical eye point-of-regard system. Oneend of the mechanical linkage is attached to a point inthe target field, and the other is held between thesubject's molar teeth, with a "pipestem" bite board(Klein & lex, Note 15). The device permits head move­ments of up to 20 em and eye point-of-regard measure­ments of better than 1 deg of accuracy over a range of20 deg vertical and 40 deg horizontal (see Figure 43).The same technique can be applied to measurement ofthe head position and orientation in all 6 deg of freedomby measuring all linkage angles and lengths. The 6 deg offreedom system allows accurate head-positionmeasurement within a volume approximately 2 ft highand 6 ft in diam, allowing full freedom in yaw, andpitch of 30 deg upward and 80 deg downward.

An additional high-accuracy mechanical head-positionsensor was developed by Vickers (Note 16) using a tele­scoping tube with one end attached to the ceiling andthe other to the subject's headset via a universal joint.The weight of the apparatus was supported by constant­tension springs. Six rotary pulse generators at the head­set measured the 6 deg of freedom changes in positionwith resolution of hundredths of an inch in translationand .3 arc min in rotation within a volume approxi­mately 2 ft high and 5 ft in diam.

Inertial head position sensors. Inertial measurementsof head motion can be made with miniature gyroscopesand accelerometers mounted on the helmets. The sizeand expense of systems to measure position accuratelyenough for eye fixation work is currently not justified.However, for gross measures of subject angular velocityor acceleration, as for studies of vestibular nystagmus,such systems are practical (Howland, 1961). When linearaccelerometers are used, care must be taken to separatethe effects of rotations of the sensitive axis into the grav­ity vector from pure linear accelerations, includingthe tangential and centripetal accelerations associatedwith rotation. Simultaneous telemetry of eye move­ments (EOG) and head linear acceleration during balletrotation is useful in a qualitative assessment of head andeye movements (Tokita, Fokuda, & Watanabe, 1973).

Electromagnetic head position sensors. Steinman(Note 17) described the use of Robinson's (1963)search coil placed on a subject's eyeglass frames todetect the angular position of his head. The subjectwas seated with sensing magnets around him, and the

EYE MOVEMENT

DEVICE~:.

..r-

Figure 43. "Pipestem" bite board device for head-positionmeasurement incorporated with an eye movement device(Courtesy Systems Technology, Inc.).

measurement was made of the relative orientation of thesearch coil with respect to the magnetic field producedby the surrounding system.

Eyeball GeometryA schematic diagram of the eyeball is given in Fig­

ure 44 for reference.

TRADE-OFFS AND GENERAL CONSIDERA·TIONS IN INSTRUMENTATION SELECTION

FOR EYE MOVEMENT RESEARCH

Trade-offsFortunately, for a particular eye movement

instrumentation requirement, not all of the highestperformance criteria of the various measurementconditions are needed to produce useful results.Clearly SOIllC parameters are more important thanothers. and the best possible system that meets theminimum requirements should be chosen andemployed (See Table 1).

As an example, a researcher may not be interestedin or need to detect fine eye movement; he may notneed to look at the high-frequency details 'of asaccade; or his range requirements may be small. Hemay therefore be satisfied with a simpler, lesspowerful instrument.

One of the first and most important trade-offs isthat of subject comfort and setup time as opposed toaccuracy and precision. One is invariably at theexpense of the other. Very high maximum accuracy ofabout .01 deg may be obtained with tight-fittingcontact lenses, but setup and discomfort shortcom­ings are obvious. Another trade-off is that of rangeand accuracy. A very high range of greater than±SO deg can be obtained with electro-oculography.That, however, is at the expense of poor de accuracy.a high percentagc of lost data, considerable setuptime, and artifact problems. Another trade-off is that

424 YOUNG AND SHEENATable I

Comparison of Eye-Movement Measuring Techniques

Measurement Accuracy Speed or Interference Classes Contact Subject Subject Subject UsableRange (deg] Frequency With Normal Acccp- Lens Variation Coopc- Train- With

Method Ver ti- Hori- Vcrti- Hori-Response Vision table Accep- Problems. ratiun inj; Rc- Young

table Eye Color. Required quircd Chil-cal zontal l'a) zontal

Etl'. ren

I. Corneal reflex (Mackworth Camera)

Polymctric ±9 ,9 .5 dog .5 deg Photo- Medium Possible N(1OC High LowLab graphic sourceVJl64 rate: 12-64 of error

frames/secTelevision:

Polymetric ±IO I deg I deg60 fields/sec

±10 Same as High: Weight No Possible None High Low NoMobile above on head; optics sourceVOl6S near eye of errorNAC· ±10- ±10- 2 deg 2 deg Same as High: Weight No Possible None High Low NoREES 20 20 above on head source

of error

lamp or Bothradiant til}- til}- Precision Highspot 30 30 3 sec 3 sec

coilLarger than IS sec IS sec High

othersmirror tlO ±10 2 sec 2 sec High

3. EOG ±50 ±SG- 2 deg L5 deg de or None80 .01-15 Hz

limited bytiltering

2. Contact lens with High No

No

Yes

No

Yes Yes

Eye must Highacceptcontact lens

Medium: Mediumplacementof electrodesand calibrationis variable

High

Low

No

Yes

4. Limbus boundaryNarco ±IO 4 deg 2 dcg 2 rnsec; Medium Yes Yes iris High Low YesEye Trac 30 msec coloration

with a factorrecorder

Nareo +lO ±20 2 deg I deg 4 msec; Medium No Yes Iris High Low YesModel -20 26 msec coloration200 with a factor

filtering

5. Wide-Angle Mackworth CameraPolymetrie 40 40 2.5 deg 2.5 deg Same as Mediurn.Subject Possible None High Low YesVII66 Method looks through source

I apertures; special of errorlighted stimuliare req uired

6. Pupil-center-corneal-reffection distanceHoneywell +30 ±30 I dcg ) deg .1 sec Low Yes Possible Low Low Low Yes

Oeulometer -10 time sourceconstant of error

Whittaker ±15 ±22 I deg J dcg 30-60 Low Yes Same as Low Low Low YesEye View Higher samples/ aboveMonitor possible sec

U.S. Army 30 40 2 dcg 2 deg 60 Low No Same as Low Low Low YesHuman sampies/ aboveEngineer- sec filtereding Lab

7 Double Purkinje

image 25 dcg 25 dcg Noise of 300 Hz. Low Autocorrcc- Low Low Low Yes

eye diam diam J min tion possible

tracker

SURVEY OF EYEMOVEMENT RECORDING METIIODS 425Table I Continued

Calibration Head Head Subject Subject Pupil lorrn of Status Cost Remark, Source ofand Setup Attach- Stabili- DIs- Aware- Diarn Output of Further

rime mcnrs zation comfort ness Out- Opera- InformationRequired Require- Put tIon

mcnt> Also

Low Chinrest High head Medium High No Photographic Commcrc..-ially High Poly metric Co.or bite- restraint or or videotape; available for P.O. Box Dboard biteboard low-resolution 111m Roseland, NJ 07068

digital output

High: Biteboard None High High No Same as above High Higher resolutionbite- lot digital output is possibleboard film with other instrumentsMedium: Head- None Medium High No Same as above High Head- Instrumcn tationfit head- mounted for mounted Marketing Corp.band, optics film TV camera 820 South Mariposaset light Burbank,CA 91506Source REES Inst. Ltd.

Westminister House, Old Woking,Surrey, United Kingdom

High: Contact Higli High No Photographic Some corn- Lens Negative

lens lens or electrical rnercial grind- pressure

must be devices ing applicationfiltered High available may may be

costly hazardousLow

Low

High: Yes, 2-6 Low Low High No Electrical Commercially Low More suit- LT Instrumentsrequires electrodes record available able for eye 4004 Osageelectrode motion than Houston, TX 77036stabiliza tion eye positionand light Grass, Inst,adaptation Quincy, MA 02169

ICS, 129 Laura Dr.Addison, IL 60101

Low Head High Low High No Analog and Commercially Low Vertical Narco BioS ystems,bracket digital available position Inc., Biometrics Div.and chinrest of eyelid 76S I Airport Blvd.

is used to Houston, TX 77017approximatevertical eyeposition

Low Spec- None Low lIigh No Analog and Cornmercrallv Low Narco (also see ICS)taeles digital available

Low Viewing Medium, Low lIigh No Photographic Commercially High Point of Polymetric Co.through head must or videotape; available for regard out- (see above)aperture be kept still low-resolution mm put without

digitizer avail head motionable artifact

Low: None Mark 1\: Low Low Yes Digital. ana Commercially Low Computer- Honeywellhigher head free. log. and fixa- available based system Radiationfor maxi- 1 in.? lion pointer Mark III 2 Forbes Rd.mum linear- Mark III: on TV image tracks head Lexington. MAization head free. of scene motion and 02172

I ft' has auto focusLow None Head free Low Medium Yes Same as above Commercially Low Tracks head Whitlllker, Space Sci.

upto available motion and Div., 33S Bear Hill Rd.1ft' has auto Waltham, MA

focus avail- 02154able

Low None I fr.' Low Low Yes Digital. ana- Research NA U.S. Army/IIELlog, video- laboratory Aberdeen Provingtape, and Ground, MDgraphic

Low Chinrestor b i tc­

board

None LowlIead freeI cm'8iteboardfor highprecision

Medium No Anal0,l!output

Smallp ro duvtion

Low HiJ\ auto Stanford ResearchfOUI\; fkld l nstf tut...'and 01"X.'r;'l- ~l'nl0 Park, CAnon are 94015dependent onpupil <izcxui table fllr image<t abjh zation

"To make measurement, not to obtain fixation point.

426 YOUNG AND SHEENA

Q. b.

"Cortex of lens and itlll capsule··Values in brackets refer to state of maximum

accommodation

Constant Eye Area or Mea8urement

Cornea 137Aqueous humor 1.33Lena capaule 1.38·

RefractiveOuter cortex, lensAnterior cortex, lens

index Posterior cortex. lenaCenter. lena 1.41Calculated total index 1.41Vitreous body 1.33

Radius of Cornea 7.7curvature, Anterior Burface, lena 9.2-122

mm Posterior Burface, lena 5 4-7.1

Distance Post. surface, cornea 1.2from Ant. surface. lens 3.5cornea, POBt. surface. lena 7.6mm Retina 24.8

FocalAnterior focal length 17.1

(142)"distance,

Posterior focal length 22.8mm (18.91

I. Focua -15.7-12.4)

Position of 2. Focus 24.4cardil"al 12 1. 0 )points I. Principal point 1.5measured 11.8)from 2. Principal point 1.9corneal (2.1)surface, I. Nodal point 7.3mm (6.51

2. Nodal point 7.6

1681

Diameter.Optic disk 2-5Macula 1-3

mm Fovea 1.5

Depth. mm Anterior chamber 2 7 - 4 2

,,-e~

SCLERA

RIGHT EYE

TEMPORAL SlOE(Nasallield)

POST ~URFAC [ l~__ .2~L1~~

PO~T ~URFACE OF CORNEA:-i---------.I

ANGLE 0' 5.7 0

..... OPTI( AXIS

II

I DIAMETER OF CORNEA 11-1) rnm II • i i • I

I 1~'f6:~~ I

: :': :I I I ANTERIOR SURFACE OF CORNEA

VISUAL A"(I~'-"

---------------------------------------,

The diagl'am and table give dimensions and optical constants of the human eye. Values in bracketsshown in the table r ef'e r to state of maximum accommodation. The drawing is a cross section ofthe right eye from above.

T'hr- horizontal and ve r t i c a l diameters of the eyeball are 24.0 and 23.5 mm, respectively. Theoptic disk, or blind spot, is about 15 degrees to the nasal side of the center of the retina andabout 1.5 degr'c,'s b cIo w the horizontal meridian.

Figure 44. Schematic and optical constants of the eyeball (Roth, Note 19; after White, Note 20; and Spector, Note 21).

between speed and noise. When video signal data orother measuring systems that inherently generatenoise are used, averaging and filtering may be used toreduce noise. but the price paid here is a loweredfrequency response.

One very important trade-off against most of theparameters is cost. One can almost always pay moreand get higher performance. This is especiallyapparent in the newer devices and systems thatemploy sophisticated optics and electronics to yieldeye-position measurement.

In reading research. for example, it may be eye

movements in themselves, the saccades, regressions.fixations, blinks. etc., that are of interest. It mayoften not be necessary to relate this information toactual points in the scene. In this case, a method likeEOG or a spectacle-mounted limbus tracker may bethe most suitable since a virtually free head istolerated and satisfactory data is obtained. If point offixation is needed, however, some head fixing ormeasuring method is needed or a point-of-regardsystem must be especially built or purchased.Performance may also be reduced in some aspects.

If testing of children is considered, then a system

SURVEY OF EYE MOVEMENTRECORDING METHODS 427

must be chosen which requires minimal subjecttraining or cooperation. This may preclude the use ofcontact lens methods or even EOG under someexperimental conditions. In these cases. aninstrument that can be set up quickly with minimumcalibration would be needed.

Considerable thought should also be given tohandling the data coming out of the eye movementmeasurement instrumentation. Again. there is atrade-oft. The simpler and cheaper the form of theoutput. the harder it is to process. A photograph or avideo tape is a simple output but is hard to handle. Ananalog signal output is the next step up. It is easier tohandle and to view but may still pose some difficultiesin large data analysis. A digital recordable output thatgoes on digital tape or into a computer is mostamenable to sophisticated analysis but is the mostcostly and difficult to interface. Also time and expenseis required for the programming to extract the desiredinformation.

Simultaneous Psychophysiological MeasurementIn addition to eye movement and position, there arc

two human eye functions from which psychologicalinferences can be drawn. and there is a great deal ofliterature available in the field. The first is blinking.Almost all of the eye movement measurementmethods will have blinks as an output artifact. It canclearly be picked out in a photograph. in an EOGrecord. or in all the systems that track the limbus. thepupil. or the corneal reflection. In methods providingan electronic output. it can clearly be seen as a pulsewith a characteristic blink duration. The secondparameter is pupil diameter. There is considerablecontroversy in the scientific community as to whetherpupil diameter is an indicator of positive and negativeemotion. with dilation indicating the former andconstriction the latter. But there is general agreement.however. that pupil dilation is an indicator ofemotional activity. whatever the sign. It is thereforevery convenient to record pupil diameter. especially ifit is part of the eye movement measurement output.and to correlate that with the part of the scene thesubject is viewing. Some of the existing commercialeye movement monitor systems give pupil diameter asan output. Koff, Elman, and Wong (Note 18) compileda comprehensive bibliography on pupillary response.

For other psychophysiological measures. someconsideration must be given to their compatibilitywith eye movement measurements. For example. ifthe required test will encumber the subject and keephim stationary. there is no value in obtaining anexpensive eye movement system that would allow freehead motion. If electroencephalography is beingperformed and electrodes are attached to the subject'shead. then EOG may be a reasonable measurementmethod since the electrodes are similar and the

procedure is being done anyway. This is, of course,assuming that it fulfills the other experiment require­ments.

It is also important to be able to simultaneouslyrecord the eve position measurement output alongwith the measures being taken in a compatibleformat where the time element is either recorded orrnav he deduced. Some marker or indication ofstimuli presentation should also be recorded. In thisway only can the various parameters being recordedbe correlated to the stimuli. Unless the effects of theinputs are strikingly apparent. statistics might beneeded to relate cause and effect where the latenciesare not known. This is one reason why digitizedoutputs amenable to computer handling aredesirable.

REFERENCE NOTES

I. YOUNG. L. R.. ZUBER. B. L.. & STARK. L.Vimal and control aspect» or saccadic f)'1' movements.NASA CR·5h4. 1%6.

2. WOLF. W .. & KNEMEYER. M. Eye movement instru­mentation, Presented at the American Education ResearchAssociation Convention. Februarv 1%9.

J. AMERIAN OPTICAL COMPANY. Rhythm reading-Theophthalmograph. Southbridge. Mass: Scientific Instruments Divi­sion. 19.17.

4. HALL. R. J., & CUSACK. B. L. The measurementof eye behuvior: Critical and selected reviews of volun­tun' 'Te movement and blinking. U.S. Army Human EngineeringLahorator\.1 echnical Memorandum 18·72. 1972.

5. MERCHANT. 1. Laboratory oculomcter, NASA CR·t422.October 1%9

h. DAVIS. P. W.. LUTZ. J. 5.. WARNER. A., &FANNIN\. A. A. Design and construction o( a portableocnlomctcr !o,. lise in trans ponution oriented human factors.11 Ii ,1/".1. Department 01 Transportation Technical Report.DOTI 5(051·71·11. 1971.

-'. HALl. R. J .. RU'FNBERG. M. A.. & MONTY. R. A.A" rxprrimentul 1J/".'stlgation 01 the visual behavior ofyOll"g heroin addicts and matched controls. U.S. ArmyHuman Engineering Laboratory. Technical Memorandum 25-73.1973.

K. PETRUK. M. W., & HUNKA, S. Oculometer laboratorysystems. lhe Canada Council Research Project S7t·890.Final Report Fcbruarv 1'174.

9. SHEEN A. D. Pattern recognition techniques for extrac­tion of features of the eye from a conventional televisionscan. In R. A. Monty. & J. W. Senders (Eds.),Ere movements and psychological processes. (Washington:National Academy of Sciences. 1976.) Hillsdale. New Jersey:Erlbaum. in press.

10. HOCHBERG. 1. Personal communication to D. Sheena, t974.II. CORNSWFET. T. N.. & CRANE, H. D. Experimental

study o( visual accnmmodation . NASA CR·2007. March 1972.12. FERRIN. F. F-4 visual target acquisition system.

Symposium on Visuallv Coupled Systems: Development and Appli­cation. Medical Divisum AMRL. Technical Report AMD-TR·73-t.1973.

IJ. HAYWOOD. J. 1.. JR. New precision electro-optical tech­nique Fir measuring pilot line of sight in aircraft coordi­dinutes. Symposium on visually coupled systems: Developmentand application. Medical Division AMRL. Technical ReportAMD·TR·7J.J. 1973.

428 YOUNG AND SHEENA

14. SAWAMURA, R. T. The ultrasonic advanced helmet-mountedsight. Symposium on Visually Coupled Systems: Developmentand Application, Medical Division AMRL, Technical ReportAMD-TR-73-1. 1973.

IS. KLEIN, R. H .. & JEX, H. R.An eye point ofregard system foruse in scanning and display research. Presented at the 15thAnnual Technical Symposium of the Society of Photo­Optical Instrumentation Engineers. Anaheim, California,September 1970.

16. VICKERS, D. L. Sorcerer's Apprentice: Head mounteddisplay and wand. PhD thesis (972), Dept. of ElectricalEngineering, University of Utah, Salt Lake City, Utah. (Seealso Air Force AMDTR No. 73-1 Symposium on VisuallyCoupled Systems: Development and Application, 1972.)

17. STEINMAN, R. M. The role or eve movements in vision andin the maintenance ot vision. Pr~se~ted at the Eye Movementsand Psychological Pro~ess Meeting, Princeton, 1974.

18. KOFF, R. H., ELMAN, D., & WONG, A. Bibliography andClassification 0.( the Literature 0.( Pupillary Response.HEW' Office of Education Contract No.OEC-6-1O-079,197\.

19. ROTH, E. M. Compendium of human responsesto the aerospace environment. NASA CR-1205(i), I, 2 November1968.

20. WIDTE, W. J. Vision. In P. Webb (Ed.), Bioastro­nautics data book. NASA-SP-300b, 1964.307-341.

21. SPECTOR, W. S. (Ed.), Handbook 0.( biological data,WADC-TR-56-273. Wright Air Dev. Center, Wright-PattersonAFB. Ohio. (AD-I 10501). October 1956.

REFERENCES

BARLOW, H. B. Eye movements during fixation. Journal ofPhysiology, 1952, 116, 290·306.

BLES, W., & KAPTEYN, T. S. Separate recording of themovement of the human eyes during parallel swing tests.Acta Dto-Laryngologica, 1973, 75, 6-9.

BOND, H. W., & Ho, P. Solid miniature silver-silverchloride electrodes for chronic implantation. Electroencephal­ography and Clinical Neurophysiology, 1970, 2&, 206-208.

BOYCE, P. R., & WEST, D. C. The frequency responsesof contact lens stalk assemblies. Vision Research, 1968,8. 475-480.

BUSWELL. G. How people look at pictures. Chicago:University of Chicago Press, 1935.

BYFORD, G. H. A sensitive contact lens photoelectric eye movementrecorder. IRE Transactions on Bio-Medical Electronics.1962, BME-9, 236-243.

CARMICHAEL, L., & DEARBORN, W. F. Reading and visualfatigue. Boston: Houghton Mifflin, 1947.

CHOUET, B. A.• & YOUNG, L. R. Tracking with head positionusing an electro-optical monitor. IEEE Transactions on Systems.Man and Cybernetics. 1974, SMC4, 192.

CLYNES, M., & MILSUM, J. H. (Eds.), Biomedical engineeringsystems. New York: McGraw-Hili. 1970.

COOK. G. Control system study 0.( the saccadic ryemovement system. ScD Thesis, Massachusetts Institute ofTechnology, 1965.

CORNSWEET, T. N. New technique for the measurement ofsmall eye movements. Journal of the Optical Society ofAmerica, 1958. 4&,808.

CORNSWEET, T. N., & CRANE, H. D. Accurate two­dimensional eye tracker using first and fourth Purkinje images.Journal of the Optical Society of America , 1973, 63, 921.

DELABARRE, E. B. A method of recording eye movements.American Journal of Psychology. 1898, 9, 572.

DITCHBURN, R. W., & GINSBORG, B. L. Involuntary eye move­ments during fixation. Journal ofPhysiology, 1953, 119, I.

FENDER, D. H. Contact lens stability. Biomedical Scienceand Instrumentation, 1%4, 2.43-52.

FINDLAY, J. M. A simple apparatus for recording micro­saccades during visual fixation. Quarterly Journal ofExperimental Psychology, 1974,26.167-170.

FORGACS, P., JARFAS, K., & KUN, L. Recording of Forgacsphotonystagmographic method. Journal ofLaryngology-Otology.1973, 87, 347-364.

GABERSEK, V. Donnees electronystagmographiques des nystagmusspontanes en derivations binoculaires et monoculaires. Revuede Laryngologie, 1969, 90, 562-575.

GEDDES, L. A., MCCRADY, J. D., & HOFF, H. E. Theimpedance nystagmogram-A record of the level of anesthesiain the horse. Southwestern Veterinarian, 1965, 19, 000-000.

GONSHOOR, A.. & MALCOLM, R. Effect of changes in illuminationlevel on electro-oculography, Aerospace Medicine. 1971.41. 138-140.

HAITH, M.. Infrared television recording and measurement ofocular behavior in the human infant. American Psychologist,1969, 24, 279-283.

HOWLAND, H. R. An angular head position indicator.S. B. Thesis, Electrical Engineering, Massachusetts Institute ofTechnology, Cambridge, Massachusetts, 1961.

ISIDKAWA, S., YAMAKAZI, A., INABI, K., & NAITO, M, Anew instrument for registering eye movement (X-Y Tracker).Acta Soc. OphthalmologyJap.. 1971, 75,1110-1117.

JEANNEROD, M., GERIN, P., & ROUGIER, J. Enregistrementbidimensional des mouvements et des pauses oculaires,a propos d'une nouvelle methode. Review of Oto­Neuro-Ophthalmology, 1966, 38, 51-53.

JONES, R. Two dimensional eye movement recording using a photo­electric matrix method. Vision Research. 1973, 13,425.

KAUFMAN. L.. & RICHARDS, W. Spontaneous fixation tendencieslor visual forms. Perception & Psychophysics, 1969, 5,85-88.

KRIS, C. Vision: Electro-oculography. In O. Glasser (Ed.),Medical physics. Chicago: Yearbook Publishers, 1960.

MACKENSEN, G. Die Geschwindigkeit horizontaler Blick­bewegungen Untersuchungen mit Hilfe der Electrooculographic.Archives of Ophthalmology, 1958, 160. 47-64.

MACKWORTH, N. F., & MACKWORTH, H. H. Eye fixationsrecorded on changing visual scenes by the television eye-marker.Journal ojthe Optical Society ofAmerica, 1958, 4&,439.

MARG, E. Development of Electro-oculogt-aphy-c--Standing poten­tial of the eye in registration of eye movement. AMAArchives ofUphthulmology, 1951, 45, 169.

MATlN. L.. & PEARCE, D. G. Three dimensional recordingof rotational eye movements by a new contact lenstechnique. Biomedical Science and Instrumentation, 1964, 2.79-95.

MERCHANT, J., & MORRISETTE, R. Remote measurement ofeye direction allowing subject motion over one cubic foot ofspace. IEEE Transactions on Biomedical Engineering. 1974.BME-21. 309-317.

MITRANI, L., YAKIMOFF, N., & MATEEFF, S. L. Combinedphotoelectric method for detecting eye movements. VisionResearch, 1972. 12. 2145.

MONTY, R. A. An advanced eye movement measuring andrecording system featuring unobtrusive monitoring and auto­matic data processing, American Psychologist, 1975, 30,331-335.

MOWRER, O. H., RUCH, R. C., & MILLER, N. E. The corneo­retinal potential difference as the basis of the galvanometricmethod of recording eye movements. American Journal ofPhysiology, 1936. 114,423.

NACHMIAS, J. Two-dimensional motion of the retinal imageduring monocular fixation. Journal of the Optical SocietyofAmerica, 1959, 49.901-908.

NA~RAC. P., MILBLED, G., PARQUET, PH.·J., LECLERCQ, M.,& DHEDIN, G. Un nouveau precede d'enregistrement desmouvements oculaires. Application aux tests de tracking.uu- Medical, 1969, 14.685-687.

PARK, R. S.. & PARK. G. E. The center of ocular rotation in

SURVEY OF EYE MOVEMENT RECORDING METHODS 429

the horh:ontal plane. American Iournal of PhysioloRY, 1933.104. 545.

RASHIIASS, C. New method for recording eye movements.Journal f!( the Optical Society of America, 1%0.SO. 642.

RASHIIASS, C., & WUTMEIMER, G. Disjunctive eye movements.JOI/"'al f!(Physio/ogy, 1%1. 159.361.

RATLIFF, F.. &RIGGS. A. Involuntary motions of the eye duringmonocular fixation. Journal of Experimental Psychologv,1950,40.687·701. .

RICHARDS, Woo & KAUFMAN, L Center-of-graviety tendenciesfor filiations and flow patterns. Perception & Psychophysics,1969. 5.81·84.

RICHTER. H. R. Principes de la photo-electronystagmographle.Review ofNeurology, 1956. 94. 138.

RIGGS, L A., RATLIFF. R., CORNSWEET, 1. c.. &CORNSWEET. T. N. The disappearance of steadily fixated test­objects. Journal o( the Optical Society or America.1953. 43, 495. .

ROBINSON, D. A. A method of measuring eye movementusing a scleral search coil in a magnetic field. IEEETransactions on Bio-Medical Electronics. 1963. RME- 10.137·145.

ROllINSON, D. A. The mechanics of human saccadic eyemovements. Journal ofPhysiology, 1964. 174.245·264.

ROBINSON, D. A. The mechanics of human smooth pursuit eyemovement. Journal of Physiology, 1965. 180. 569·591.

SALAPATEIt, P., & KESSEN, W. Visual scanning of trianglesby. the human newborn. Journal of Experimental ChildPsychology, 1966,3,155-167.

SCHOTT, E. Uber die Registrierung des Nystagmus undanderer Augenbewegungen vermittels des Seitengalvenometers.Deutsches Archiv.lur Klinische Medizin. 1922. 140. 79-90.

SHACItEL, B. Review of the past and present in oculography.Medical electronics proceedings or the second internationalconference, London: Hiffe, 1%0. 57.

SHACKEL, B. E~ movement recording by electro-oculography.In P. H. Venables & I. Martion (Eds.), A manualof psychophysiological methods. Amsterdam: North-HollandPublishiOilce., 1967. Pp. 300-334.

SHEENA, D. A system for instantaneous wide angle eye posi­tion measurement. PhD Thesis, Massachusetts Institute ofTechnology, 1%9.

SHEENA, V. Two digital techniques for eye position measurement.Presented at the Eighth Annual Meeting of the Associationfor the Advancement of Medical Instrumentation. Washington.D.C.. 21·24 March 1973. Medical Instrumentation. 1973. 7. 78.

SMITH. W. M.. & WARTER. P. 1. Eye movement and stimulusmovement: New Photoelectric electromechanical system forrecording and measuring tracking motions of the eye.Journal C!( the Optical Society ofAmerica, 1%0. SO. 245.

STARK, L.; & SANDBERG. A. A simple instrumentfor measuring rye movements. Quarterly Progress Report 62.Research Laboratory of Electronics. Massachusetts Institute ofTechnology. 268. 1%1.

STARK, L., VOSSIUS, G., & YOUNG. L. R. Predictive controlof eye tracking movements. IRE Transactions on HumanFactors in Electronics, 1%2. HFE-3. 52.

STEER. R. W.. JR. The intluence of angular and linear accelerationand thermal stimulation on the human semicircular cana..ScD thesis. Massachusetts Institute of Technology. August 1%7.

SUllIVAN. G.. & WELTMEN, G. The impedance oculogram-Anew technique. Journal of Applied Physiology. 1963, i8.215-216.

TAYLOR. S. E.. The dynamic activity of reading: A modelof the process EDL Researcn and Information Bulletin.New York: McGruw-Hill. 1971. P. 9.

THOMAS. E. Ll.EWELLYN. Movements of the eye. ScientificAmerican. I%li. 219.88-95.

TOGlIA. J. 0. Electronystagmography. Diseases of the Nervous"'~\,stem. 1973. 34. 379-383.

TOKITA. T. FOKUDA. T.. & WATANABE, T. Telemeteringof eye and head movements in ballet rotation. Aggressologie,I'm. 6. 21-27.

TOROK, N.. GUIl.lEMIN, V., & BARNOTHY, J. M. Photoelectricnystagmography. Annals of Otology. Rhinology and Laryngol·ogy. 1951, 60. 917.

UENOYAMA. K .• UENOYAMA, N., & hNUMA, I. Vector­electroculography and its clinical applications. British Journalo] Ophthalmology. 1964, 48. 318-330.

VOSSIUS. G. Das System der Augenbewegung. Zeitschrifttur Biologie, 1%0, 112. 27.

WESTHEIMER. G. Eye movement responses to a horizontallymoving visual stimulus. AMA Archives ojOphthalmology, 1954,52. 912.

WHEELESS. L. L.. JR., BOYNTON. R. M., & COHEN. G. H. Eyemovement responses to step and pulse-step stimuli. Journal0/ the Optical Society otAmerica, 1966. 56, 956-960.

. WOELLNER, R. c.. & GRAYBIEL. A. Counterrolling of the eyesand its dependence on the magnitude of gravitational orinertial force acting laterally on the body. Journal of AppliedPhysiology. 1959. 14.632-634.

YARBUS. A. L. Eye movements and vision. New York:Plenum Press. 1%7.

YOUNG. L. R. The current status of vestibular system models.Automatica, 1969. 5. 369-38.3.

YOUNG. L. R. Recording eye position. In M. Clynes &J. H. Milsum (Eds.). Biomedical engineering systems.New York: McGraw-Hill. 1970.

ZUBER, B. L. Physiologicul control of eye movements in humans.PhD thesis. Massachusetts Institute of Technology, 1965.

ZUBER. B. L. & STARK. L. Dynamical characteristics of thefusional vergence eye-movement system. IEEE TransationsSystems. Science and Cybernetics. 1968. SSC-4, 72-79.

NOTES

1. From Young. 1970.

(Received for publication AUlust S. 1117.;revision accepted MaY 5. 11175.)