Ultrasound for Gaze Estimation—A Modeling and Empirical Study

Sensors

Sachin S. Talathi¹  Andre Golard¹ 

[1] Facebook Reality Labs, Redmond, WA 98052, USA;

关键词: eye tracking;    gaze estimation;    ultrasound;    CMUT;    machine learning;    Gradient Boosted Regression Trees;   

DOI  :  10.3390/s21134502

来源: DOAJ

【 摘 要 】

Most eye tracking methods are light-based. As such, they can suffer from ambient light changes when used outdoors, especially for use cases where eye trackers are embedded in Augmented Reality glasses. It has been recently suggested that ultrasound could provide a low power, fast, light-insensitive alternative to camera-based sensors for eye tracking. Here, we report on our work on modeling ultrasound sensor integration into a glasses form factor AR device to evaluate the feasibility of estimating eye-gaze in various configurations. Next, we designed a benchtop experimental setup to collect empirical data on time of flight and amplitude signals for reflected ultrasound waves for a range of gaze angles of a model eye. We used this data as input for a low-complexity gradient-boosted tree machine learning regression model and demonstrate that we can effectively estimate gaze (gaze RMSE error of 0.965 ± 0.178 degrees with an adjusted $R^2$ score of 90.2 ± 4.6).

【 授权许可】

Unknown