Estimating Gaze From Head And Hand Pose And Scene Images For Open-Ended Exploration In Vr Environments

The widespread utility of eye tracking technology has created a growing demand for more consistent and reliable eye-tracking systems, and there is a need for new and accessible approaches that can enhance the accuracy of eye-tracking data. Previous studies have offered evidence for associations between certain non-eye signals and gaze such as a strong coordination between head motion and gaze shifts. e.g. [3] , hand and eye spatiotemporal statistics, e.g. [7] , and gaze behavior and scene content, e.g. [2] . Previous studies have also shown how various combinations of eye, head, scene, and hand signals can be leveraged for applications such as gaze estimation [5] , [10] , prediction [8] , and classification [6] . Though these previous approaches provide support for the idea that non-eye sensors (i.e. head, hand, and scene) are useful for estimating gaze, they have not yet fully addressed how these signals individually and in combination contribute to gaze estimation.