Identifying gaze gestures from noisy image-based eye movement data

Abstract

This work reported here hence builds on the idea of gaze gesture identification for HCI, previously proposed in the literature, by devising robust eye-gesture recognition techniques that perform reliably on noisy image-based eye movement data obtained from a passive eye-tracker based on a single low-cost off-the-shelf web-camera. In our work, we define gestures as a sequence of trajectories generated as the user fixates upon a number of gaze markers created purposely for the task. Reliable identification of different gaze gestures requires the extraction of fixation periods from the eye movement data. Hence, we propose an online Kalman filter-based algorithm that reliably demarcates fixation periods in noisy image-based data. This allows us to identify fixation sequences and the trajectories between them for gaze gesture recognition, and permits the design of a gesture vocabulary that may be reliably recognised from our low-cost platform.