Applied Science Laboratories only uses performance specifications that correspond to actual use. System performance, when measured under artificial conditions, will be better than that stated but most likely will never be achieved when working with human subjects. Please keep this in mind when evaluating any eye tracking system.

ASL uses the following terms to define performance for its eye trackers:

Accuracy and Precision:
As a general concept, accuracy is the expected difference between true line of gaze and measured line of gaze and is usually expressed in terms of visual angle. Is is a measure of how good the absolute eye position measure is, with respect to the room in the case of a remote system or with respect to the head in the case of a head mounted system.

There are many different specific ways this concept may be defined. An operational definition often used by ASL is the expected difference in degrees of visual angle between true eye position and mean computed eye position during a fixation. Precision can be though of as the amount of instrument noise (jitter) in the eye position measure when the eye is perfectly stationary. It can be described as the standard deviation of the measurement, in degrees visual angle, when the eye is stationary.

Note that a real eye is never perfectly stationary and when people fixate a spot, they do not center it on the fovea with infinite accuracy. If a real eye is used to measure eye tracker accuracy and precision, this physiological uncertainty must be taken into account.

Be wary of tests using model (artificial) eyes. It is extraordinarily difficult to duplicate all properties of a real eye and performance achieved with a model eye may be misleading.

Calibration:
Different geometrical configurations (i.e. relative position of subject, optics and scene) and between-subject differences must be accounted for to achieve accurate and/or linear eye line of gaze measurement for all subjects and conditions. Calibration is the process or procedure for removing the effect of these subject and set-up differences. A calibration procedure usually requires the subject to look at a certain number of predetermined target points. The relation between the raw measured value (e.g. separation between pupil and corneal reflection image) and the final device output is adjusted manually or automatically by a computer. It is very advantageous for the calibration procedure to be quick and easy for both the equipment operator and the subject.

Calibration can vary both procedurally and mathematically among different eye tracking devices, and these differences may have profound effects on system accuracy, linearity and ease of use.

Compensatory Eye Movements:
Compensatory eye movements are conjugate eye motions that partially stabilize the visual field during either active or passive head or trunk motions. Such eye motions are involuntary and are in response to inertial information from the vestibular system (located in the inner ear) and head motion information supplied by proprioceptive sensors in the neck.

Corneal Reflection:
Also called the first Purkinjie image, the corneal reflection or corneal reflex, is the reflection of a light source on the outer surface of the cornea.

Fixations and Saccades:
During normal scanning of a visual scene, eye movement is characterized by a series of stops and very rapid jumps between stopping points. The stops, always lasting at least 100 msec, are called fixations. It is during these fixations that most visual information is acquired and processed. The rapid jumps between fixation points are called saccades. Saccades are conjugate eye movements (both eyes move together) that can range from 1 to 40 degrees visual angle, generally have durations of 30 to 120 msec, and achieve velocities as high as 400-600 deg/sec. Very little visual information is acquired during saccades, primarily because the very fast motion of images across the retina and partially because of an elevated visual threshold just prior to and during a saccade.

The eyes are not completely stationary during fixations. Rather they exhibit a variety of small involuntary motions, usually of less than one degree visual angle magnitude called flicks, drifts and tremors. Smooth pursuit, compensatory eye movement, vergence, and nystagmus are nonsaccadic eye movements of relatively large magnitude.

Flicks, Drifts and Tremors:
At least three types of small, involuntary eye motions commonly occur during eye fixations. Flicks are very rapid (perhaps as little as 30 msec apart), involuntary, saccade-like motions of less than 1 degree. Drifts are very small, slow (about 0.1 deg/sec), apparently random motions of the eye. Tremors are tiny (less than 30 arc seconds), high frequency (30-150 Hz) eye vibrations.

Fovea:
The fovea is a small, roughly disk shaped area on the retina that offers the highest visual acuity. It is located about 2 mm (corresponding to about 5 degrees visual angle) to the temporal side of the central optic axis (line through the center of the cornea and lens). The diameter of the fovea corresponds to about 2-3 degrees visual angle. In other words, if a person looks at a disk with a diameter that subtends one degree visual angle, the image of the disk that is focused on the retina will cover an area about equal to that of the fovea. The word is sometimes used as a verb, to foveate, meaning to position the eyes so that the retinal image of a certain target or element of the visual scene falls on the fovea. In fact to "fixate" a point in the visual field implies foveation of that point. When an eye tracking instrument measures the point of fixation on a scene, it is presumably indicating the part of the scene being imaged on the fovea.

Linearity:
Linearity is the degree to which a change in device output is proportional to change in eye position. Stated another way, it is the degree to which a plot of device output versus true eye position would be a straight line.

It is generally specified as a percent of the size of the excursion. If linearity is 10%, for example, an eye motion of 10 degrees visual angle should be measurable to within a degree, a 20 degree excursion to within 2 degrees, and so on.

Note that a device can have a large offset error and still be linear as long as changes in output remain proportional to changes in eye position.

Nystagmus:
Nystagmus is an involuntary sawtooth pattern of conjugate movement that occurs in response to apparent motion of the visual field (especially the peripheral field) or inertial rotation of the body. Each cycle is characterized by a "slow phase" when the eyes move to stabilize the visual field on the retina, followed by a return saccadic jump or "fast phase." The slow phase velocity and frequency of the pattern are related to the speed of the visual field motion or the speed of head rotation up to a maximum nystagmus frequency of about 5 Hz. The amplitude is generally between 1 and 10 degrees visual angle. Onset and changes in nystgmus slow phase velocity are often similar although not identical to a personís perception or sensation of angular rotation velocity.

Precision:
The amount of instrument noise (jitter) in the eye position measurement when they eye is stationary.

Resolution:
Resolution specifies how finely the output scale is divided or, in other words, the smallest change that can ever be observed in the measured value. Resolution is introduced by the digitization of a process and has no meaning for analog processes.

Smooth Pursuit:
The eyes can smoothly track targets that are moving in the range of 1 to 30 degrees per second. These conjugate, slow tracking eye movements are usually called smooth pursuit and act partially to stabilize slowly moving targets on the retina. Slow, smooth eye movements cannot generally be executed voluntarily without a slowly moving target. Voluntary control, however, has been demonstrated after training.

Transport Delay:
Transport delay specifies the temporal relationship of each data sample to the actual event being measured. For example, a system might provide 1000 data samples per second but the entire data stream might be delayed by one second. Thus an event at t=0 sec would be delayed by one second.

For applications that use eye position data to control or switch important. If data, however, are to be used only for off-line analysis, transport delay does not matter at all since it can simply be subtracted out. Transport delay does not affect the frequency content of the information at all, it just determines when the information is received.

Update Rate:
Measurement update rate specifies the number of independent data samples per second provided by the instrument. Update rate will affect the information content. A slow update rate will cause high frequency information to be irretrievably lost.

Vergence:
Vergences eye movements are non conjugate eye movements needed to keep the visual scene in the same relative position on both retinas. For example, if a person is fixating (foveating) a target point and the target begins moving closer to the person, the eyes must rotate towards each other (both must rotate towards their nasal side) to maintain the image on both foveas and thereby retain fusion of the image. Vergence eye movements have a range of about 15 degrees visual angle and maximum velocities of about 10 degrees per second.

Visual Angle:
Visual angle refers to the angle between any two vectors that have their origin at the eye. Assuming the head remains stationary, the amount that the eye has to rotate to foveate two different targets (the angle between the vectors connecting the eye to each target) can be referred to in terms of visual angle. Visual angle is often specified in degrees, arc minutes, or arc seconds. There are 60 arc minutes to a degree and 60 arc seconds to an arc minute.