Calibration of gaze tracking equipment

Summary of third fixation study, comparison of animated and static fixation targets at kings Buildings Edinburgh 1997
J S Strachan.

Apparatus:
Gadarian gaze tracking system
Dynamic differential IR scleral sensor (DDIS)
Computer animated targets.
Various real world static and moving targets





Method
A differential IR sensor placed close to the sclera was used to determine the stillness of the eye at a given fixation. This method can detect motion of less than 3 seconds of arc but is subject tp serious drift and so is not used for gaze direction data.

Relative gaze direction was measured using the Gadarian bright pupil occulography eye tracker. This was used to determine relative gaze direction only. The purpose of the experimental series is to determine the most reliable fixation targets for calibration of absolute gaze direction measurement.

A sequence of targets was presented to 14 subjects 12 male and 2 female ranging in age from 22 to 38. The stability of the fixation on each target was measured with the DDIS and the timing and choice of fixation targets was monitored by the Gadarian tracker.

Targets included text as individual words and letters, recognizable organic targets real and computer animated insects, a captured video sequence of a human eye and a still image of a human eye. Also included were common fixation objects such as clock faces of various styles. Each target was presented in several sizes covering the range of 0.5 degrees to 3 degrees.
In addition to the objective measurements the subjects were interviewed to determine if they were aware of their relative fixation stability

Conclusion
Fixation stability on static targets was generally as reported in previous studies decaying from maximum stability of 4 arc seconds at 200mS to 600 mS gradually decaying to 1 to 2 degrees by 20 seconds. Fixation beyond one minute varied dramatically between subjects with only 30 percent (4 subjects) capable of maintaining a sub 5-degree fixation reliably for a minute. Text in the form of single words provided the worst fixation results except where the word was dimensionally "center weighted" such as "fallen" "eye" "the".
Of the dynamic targets, any target which was less than 2.5 degrees in diameter which had an animated perimeter resulted in substantial improvement in long term fixation stability improving the 1 and 2 minute fixation stability to better than 30 seconds of arc. This represents an improvement of more than a factor of ten in the best subjects performance but of even more interest was the fact that none of the subjects failed to maintain a 1-minute fixation of a degree or better. Of the animated targets presented a computer animated beetle with wiggling legs presented at 2 degrees diameter was statistically the best fixation target in the test series. 100 percent of the subjects maintained fixation to better than 1 degree for five minutes and 100 percent of subjects maintained a fixation of better than 10 seconds of arc for 20 seconds.

While the "wiggly legged bug" target was by far the best for maintenance of a fixation it was not the best as a stimulus for maximum accuracy of saccade. In that case the animated human eye target excelled as the most "attractive" target to stimulate a single ballistic saccade.

Close analysis of the spatial frequency of each target and the progress of fixation drift indicates a possible explanation for both fixation drift and the efficacy of the wiggly-legged bug target. Fixation targets with inherently low spatial frequency i.e. soft edged targets produced by far the largest overall drifts with errors as high as 10 degrees before correction. Targets of high spatial frequency tend to cause corrections more often at least in the initial 1 to 2 minute period. The dynamic perimeter targets such as the wiggly-legged bug appear to continuously stimulate the fovea so that it tends to settle on the point that balances the dynamic spatial frequency of the animated portion of the target around the foveal perimeter. It is suggested that consciousness of fixation drift depends on sufficient stimulation of the fovea.

It is worth noting that there was virtually no correlation between a subjects belief in their performance or quality of fixation and actual objective measurement. Encouragement and reminders help to maintain fixations to within five degrees but below that it appears that subjects are not aware of the drift and cannot consciously improve it at will.