Tobii TX300 Eye Tracker
17
Accuracy and precision measurements
Gaze accuracy and precision are typically measured in degrees of visual angle. One degree accuracy
corresponds to an average error of 12 mm (0.47’’) on a screen at a distance of 65 cm (27’’). Data is
presented as monocular or binocular. Monocular data is based on data from the subject’s dominant eye
only. Binocular data is the average of both eyes.
Gaze precision
Precision measurements are done using artificial eyes to eliminate artifacts from human eye movements.
Tobii specifies precision both with and without noise reduction filters. All measurements are done at the
default sampling rate of the eye tracker and at the optimum distance between the eye tracker and the
subject. For TX300, the sampling rate was 300 Hz and the distance 65 cm (27’’). Precision is calculated
as root-mean-square (RMS) of successive samples.
Using a noise reduction filter can improve precision and it is often used when doing eye tracking analysis.
For comparison, Tobii provides precision values both with and without applying a filter. The filter used
is a Stampe stage 2 algorithm. More information about this filter can be found in Behavior Research
methods, Instruments & Computers 1993, 25 (2), pp. 137-142.
As the distance from the eye tracker influences precision, measurements are also taken at various
distances. Data from these measurements is binocular.
Gaze accuracy
Accuracy under ideal conditions is measured in the center of the head movement box with the subject
fixed in a chinrest. Data is collected immediately after calibration, in a controlled laboratory environment
with constant illumination, with 9 stimuli points at gaze angles ≤18º. Measurements are done on 20 test
subjects without lenses, glasses or droopy eyelids. Accuracy for one subject is calculated as the mean
of several data samples for several stimuli points across a screen. The accuracy figure presented is the
mean accuracy from all subjects.
Good accuracy is difficult to achieve at large gaze angles, but is important when testing large stimuli.
For instance, the upper corner of a 23’’ screen with the test subject at a distance of 65 cm (27’’) from
the eye tracker corresponds to a 31º visual angle from the center of the eye tracker unit. Consequently,
measurements are also presented for stimuli presentations at large gaze angles.
When luminance of the stimuli or the illumination in the lab changes, the size and shape of the pupil
is affected. Unless compensated for, this may cause a significantly reduced accuracy. Testing the
influence of surrounding light and stimuli luminance on accuracy is done in a laboratory environment
with controlled light conditions. Stimuli points are presented on a black background so as not to influence
ambient light conditions. Testing is also done with ambient light at a level deemed ‘normal’ office lighting
where the background is changed to white with black stimuli points. The results from a test in which the
background was changed has in previous Tobii product descriptions been referred to as Drift.
As with precision, the distance from the eye tracker influences accuracy. When testing this influence,
calibration is done with the subject in the center of the head movement box (i.e. at a distance of 65 cm)
and measurements are done thereafter. Measurements are performed with the test subject at precise
and specific distances relative to the eye tracker, measured along the axis of the tracking sensors. Data
from these measurements is binocular.
Appendix II: Measurements
To Next Page
Loading...
