Absolute Eye Gaze Estimation With Biosensors in Hearing Aids.
Level 4 - case-series / case-control
Small exploratory within-subject laboratory validation study without a comparison control group.
PubMed 31920477 · doi:10.3389/fnins.2019.01294
What was done
Eleven hearing-impaired participants took part in a simulated multi-talker listening task with targets positioned at -30°, 0°, and +30° azimuth. Participants tracked visual LED targets across two conditions: head-fixed (using a chinrest) and head-free. In-ear electrodes captured electrooculography (EarEOG) signals alongside ear-mounted motion sensors. Researchers modeled relative and absolute eye-gaze estimation while accounting for saccades, fixations, head movement, and electrode-skin half-cell drift.
What was found
The EarEOG model explained 90.5% of the signal variance when the head was fixed and 82.6% when the head moved freely. Absolute eye-gaze estimation was reliable in the head-fixed condition. However, hardware issues introduced excessive variance in the head-free condition, preventing reliable identification of the attended target.
Why it matters
This work shows the potential of ear-level biosensors to detect visual attention for steering directional microphones in hearing aids, while highlighting technical hurdles that remain for free-head real-world use.
Limits
The sample size was small (n = 11), testing was restricted to three discrete angles in a controlled lab setting, and hardware constraints directly prevented reliable gaze estimation during realistic head movement.