Ultrasound for Gaze Estimation-A Modeling and Empirical Study

Area of Science:

• Biomedical Engineering
• Sensor Technology
• Human-Computer Interaction

Background:

• Conventional light-based eye tracking methods struggle with ambient light variations, limiting outdoor applications, particularly for augmented reality (AR) glasses.
• Ultrasound sensing presents a promising, low-power, and light-insensitive alternative for eye tracking, overcoming limitations of camera-based systems.

Purpose of the Study:

• To model the integration of ultrasound sensors into an AR glasses form factor for eye-gaze estimation.
• To evaluate the feasibility of using ultrasound for gaze tracking in various configurations.

Main Methods:

• Developed a model for integrating ultrasound sensors into AR glasses.
• Designed a benchtop experimental setup to collect time-of-flight and amplitude data of reflected ultrasound waves from a model eye across different gaze angles.
• Utilized a gradient-boosted tree machine learning regression model with empirical data for gaze estimation.

Main Results:

• Demonstrated effective eye-gaze estimation using ultrasound signals.
• Achieved a gaze root-mean-square error (RMSE) of 0.965 ± 0.178 degrees.
• Obtained a high adjusted R-squared score of 90.2 ± 4.6, indicating strong model performance.

Conclusions:

• Ultrasound technology is feasible for eye-gaze estimation in AR glasses.
• The developed low-complexity machine learning model effectively processes ultrasound data for accurate gaze tracking.
• This approach offers a viable, light-insensitive solution for eye tracking in diverse environments.