Abstract
Some video-based eye tracking techniques rely on estimating pupil center to measure eye movements. However, estimation of pupil center can be distorted by variations in pupil size, which are known to reflect changes in retinal illuminance, arousal and task demands. Earlier studies have shown that these distortions can create spurious changes in the measurement up to several degrees in visual angle, thus weakening the validity of results reliant on accurate estimates of eye position. We have developed and validated a regression-based method for correcting this pupil size artifact. In a visually guided saccade task, twenty-three observers either maintained fixation on a dot (diameter 0.12°) located in the center of a monitor screen (fixation condition) or tracked the small dot as it jumped 0.12° leftward or rightward from the center at pseudorandom times (saccade condition). Eye positions and pupil size were sampled binocularly at 500 Hz using a video-based eye tracker (EyeLink 1000, SR Research). The task allowed us to distinguish actual eye movements, including microsaccades, from spurious eye movement signals associated with pupil size changes. Confirming previous reports, we found high correlations between measured gaze positions and pupil size, which accounted for 43.6% ± 29.5% (mean ± SD) and 21.3% ± 25.8% of the variance in the left and right horizontal eye traces, respectively, and 35.6% ± 27.6% and 35.0% ± 27.7% of the variance in the left and right vertical eye traces, respectively. Notably, the pupil-confounded mean gaze positions of the central dot in the fixation condition deviated substantially from those of the other fixation targets in the saccade condition (absolute deviation=0.22°±0.14°), but after correction the deviation was reduced significantly (0.12°±0.09°; paired t-test p=0.001). Our findings confirm that the inherent imprecision of pupil-based eye tracking can be effectively mitigated thereby providing more accurate, reliable eye movement measurements.
Meeting abstract presented at VSS 2014