IR Light Sensing at the Edges of Glasses Lenses for Invisible Eye Tracking

We propose a novel readout configuration of photo detectors (PD) employed for infrared (IR) oculography placed on the lateral edges of the standard lenses of glasses for eye tracking. They collect the amount of light reflected by the IR-illuminated eyeball and captured by the lens, which guides it towards the edges. A set of discrete photodetectors is thus hidden inside the rims and their signals are fed to an agile artificial neural network (ANN) allowing to reconstruct the direction of the incident light, thus enabling the estimation of the direction of gaze. Here we report the experimental validation of this sensing approach on both a rectangular and prescription lens. The accuracy of the ANN inference applied to the current preliminary setup with 4 PDs is ~5° and ~25° for pitch and yaw respectively. The robustness of the detection was also tested under spurious IR illumination, which is fully rejected by a differential measurement. We believe that this approach will allow the combination of aesthetics, suitable accuracy and low-power needed to shift eye tracking from specific trial conditions and devices (such as headsets with bulky external batteries) to a pervasive, full-time feature of next-generation smart glasses.