A hybrid high speed eye tracking sensor by fusion of video oculography and laser feedback interferometry

High-speed eye tracking is a key requirement for upcoming AR and VR devices, as it enables novel applications such as display enhancement through gaze-contingent rendering, saccadic endpoint prediction to minimize system latency, and gaze-based biometric user identification. The update rate of state-of-the-art mobile video oculography sensors is limited by system-level power consumption, which is mainly caused by the exponentially increasing computational power requirements for pupil detection in camera images with respect to the update rate. To overcome this limitation and enable unconstrained high-speed eye tracking with low power consumption, we propose to fuse eye movement velocity data acquired by laser feedback interferometry sensors with camera images acquired at a low sampling rate. We propose a model-based sensor fusion approach that involves a combination of a glint-free, model-based video-oculography approach and high-speed interpolation between successive camera images using velocity information derived from the laser feedback interferometry sensors to achieve an outstanding eye tracking sensor system update rate of 1 kHz. We evaluated the proposed hybrid sensor system in an experimental setup and achieved a gaze accuracy of 1.63°.