Scalable Eye Tracking for Mobile Devices

Abstract

Eye tracking provides an effective solution to users’ attention, interest, and engagement. While gaze estimation based on a standard camera can be versatile, it remains challenging to achieve an accurate, robust, and scalable solution on mobile devices. In this talk, I will describe three studies that aim to address these challenges. Specifically, 1) we found that screen reflection on user's cornea can be leveraged for gaze estimation and it considerably improves the practicability of indoor eye tracking. 2) We exploited gaze-hand coordination and applied interaction data for implicit calibration when a user naturally interacts with the computer. This can prevent users from tedious and intrusive calibration in practice. 3) We also proposed to train a multi-device person-specific gaze estimator to accelerate implicit calibration. It adapts the data from different personal devices to learn the shared mapping from user appearance into eye gaze. Taken together, these studies identify indicative eye gaze features, alleviate user calibration effort, and thus pave the way for scalable eye tracking in daily use.