Gaze estimation using 3-D eyeball model and eyelid shapes

Abstract

This paper proposes a gaze estimation algorithm using 3-D eyeball model and eyelid shape. The gaze estimation suffers from differences of eye shapes and individual behaviors, and requires user-specific gaze calibration. The proposed method exploits the usual 3-D eyeball model and shapes of the eyelid to estimate gaze without user-specific calibration and learning. Since the gaze is closely related to the 3-D rotation of eyeball, this paper first derives the relation between 2-D pupil location extracted in the eye image and 3-D rotation of eyeball. This paper also models the shapes of the eyelid to adjust gaze based on the observation that the shapes of the eyelid are deformed with respect to the gaze. This paper models the curvature of eyelid curve to compensate for the gaze. According to the various experiments, the proposed method shows good results in gaze estimation. The proposed method does not need user-specific calibration or gaze learning since the general 3-D eyeball and eyelid models are exploited in the localized eye region. Therefore, it is expected that the proposed gaze estimation algorithm is suitable for various applications such as VR/AR devices, driver gaze tracking, gaze-based interfaces, and so on.