Accurate and Robust Eye Tracking with Ultrasound: A Computational Study

This paper presents an ultrasound simulation platform to synthesize realistic ultrasound eye tracking data as a function of transducer/ system design, sensor noise, eye/ face occlusion, and headset slippage. Simulation data were synthesized using a single face with adjustable gaze and eyelid opening in the presence of headset slippage. The data generated using this face/eye model was input into a machine learning algorithm to jointly estimate gaze and headset slippage. We achieved gaze root-mean-square-error (RMSE) of 0.085° and 0.756° without and with headset slippage, respectively. We anticipate that the proposed end-to-end simulation pipeline will enable tractable design optimization of wearable ultrasound devices and facilitate further investigation of ultrasound sensing solutions as a complementary technology to camera-based eye-tracking for AR/VR applications.