Gaze tracking accuracy in humans: One eye is sometimes better than two

Area of Science:

• Ophthalmology
• Computer Science
• Human-Computer Interaction

Background:

• Modern video eye trackers provide binocular data, enabling the calculation of a version signal by averaging left and right eye data.
• Averaging eye signals is a common practice to enhance data precision (reduce variable error).

Purpose of the Study:

• To investigate the impact of averaging binocular eye signals (version signal) on systematic error (accuracy).
• To compare the accuracy and precision of individual eye signals versus the averaged version signal.

Main Methods:

• Two calibration validation experiments were conducted with a total of 143 participants (n=79 and n=64).
• Systematic error was calculated independently for the left eye, right eye, and version signals for each participant.

Main Results:

• In 29.5% (first experiment) and 25.8% (second experiment) of participants, the systematic error of a single eye signal was lower than that of the version signal.
• Using the version signal can lead to a higher systematic error compared to individual eye signals in a significant portion of users, albeit with improved precision.

Conclusions:

• The choice between using individual eye signals or the version signal depends on the research priority: precision or accuracy.
• For studies prioritizing precision, averaging binocular data (version signal) is recommended.
• For studies prioritizing accuracy, selecting the individual eye signal (left, right, or version) with the lowest systematic error is advised, even if it compromises precision.