Natural human-robot interaction requires leveraging viewing direction information in order to recognize, respond to, and even emulate human behavior. Knowledge of the eye gaze and point of regard gives us insight into what the subject is interested in and/or who the subject is addressing. In this paper, we present a novel eye gaze estimation approach for point-of-regard (PoG) tracking. To allow for greater head pose freedom, we introduce a new calibration approach to find the 3D eyeball location, eyeball radius, and fovea position. To estimate gaze direction, we map both the iris center and iris contour points to the eyeball sphere (creating a 3D iris disk), giving us the optical axis. We then rotate the fovea accordingly and compute our final, visual axis gaze direction. Our intention is to integrate this eye gaze approach with a dual-camera system we have developed that detects the face and eyes from a fixed, wide-angle camera and directs another active pan-tilt-zoom camera to focus in on this eye region. The final system will permit natural, non-intrusive, pose-invariant PoG estimation in distance and allow user translational freedom without resorting to infrared equipment or complex hardware setups.
Mendeley saves you time finding and organizing research
Choose a citation style from the tabs below