Exponentially smoothed interactive gaze tracking method

Abstract

Gaze tracking is an aspect of human-computer interaction still growing in popularity. Tracking human eye fixation points can help control user interfaces and eventually may help in the interface evaluation or optimization. Unfortunately professional eye-trackers are very expensive and thus hardly available for researchers and small companies. The paper presents very effective, exponentially smoothed, low cost, appearance based, improved gaze tracking method. The method achieves very high absolute precision (1 deg) at 20 fps, exploiting a simple HD web camera with reasonable environment restrictions. The paper describes results of experimental tests, both static on absolute gaze point estimation, and dynamic on gaze controlled path following.