HeadGaze: Natural Gaze Estimation While Walking From a Single IMU on Earbuds

Gaze provides valuable insights into people’s interests and can facilitate seamless interaction with computational systems. However, capturing natural gaze during dynamic motions with compact, low-cost, low-power wearables like earbuds is challenging. While gaze estimation using IMUs has been explored, it typically requires the user to remain still or relies on additional data such as posture, limiting its use in various scenarios. In this work, we propose HeadGaze, which estimates gaze also in walking postures only from a single IMU inside an earbud with an LSTM network. We hypothesized that the acceleration caused by walking can be utilized to estimate the head orientation. To estimate gaze from head orientation, we obtained the linear head-gaze relationship coefficients. Our user study, in which participants looked freely in various postures, showed that our method estimated gaze relative to the body with an error of 12.9 degrees in the yaw direction and 10.2 degrees in the pitch direction. HeadGaze enables gaze estimation using a single IMU inside a wearable device, such as earbuds and smartglasses worn on the head, making gaze interaction techniques available on various devices. Moreover, our method essentially requires only earbuds and a smartphone for user-specific data collection without requiring special equipment such as eye trackers.