Noise estimation for head-mounted 3D binocular eye tracking using Pupil Core eye-tracking goggles

Area of Science:

• Biomedical Engineering
• Computer Vision
• Human-Computer Interaction

Background:

• Head-mounted video-based eye tracking systems are increasingly utilized across various applications.
• Accurate gaze estimation is crucial for reliable data in research and development.
• Understanding measurement uncertainty is vital for robust experimental design.

Purpose of the Study:

• To systematically assess sources of measurement uncertainty in the Pupil Core eye tracker.
• To evaluate uncertainty across 2D image, 3D eye rotation, and external gaze projection.
• To investigate the impact of eye camera motion on gaze alignment and data quality.

Main Methods:

• Practical assessment of measurement uncertainty for the Pupil Core device.
• Analysis of uncertainty in 2D scene camera imaging.
• Evaluation of 3D eye rotation and external gaze point projection accuracy.
• Assessment of eye camera motion during active tasks relative to eye and scene cameras.

Main Results:

• Eye camera motion, inaccurate gaze point depth estimation, and flawed eye models introduce significant noise.
• Calibration accuracy and precision may not fully reflect gaze point estimation errors.
• Variability in gaze point estimation can be substantial and requires careful consideration.

Conclusions:

• Eye model constancy is important for valid comparisons across experimental conditions.
• Additional assessments of data reliability are recommended for experiments involving gaze point estimation or external world-relative eye movements.
• Researchers must account for identified uncertainty sources in experimental design for accurate eye-tracking data.