Non-intrusive eye gaze tracking under natural head movements

We propose an eye gaze tracking system under natural head movements. The system consists of one CCD camera and two mirrors. Based on geometric and linear algebra calculations, the mirrors rotate to follow head movements in order to keep the eyes within the view of the camera. Our system allows the subjects head to move 30 cm horizontally and 20 cm vertically, with spatial gaze resolutions about 6 degree and 7 degree, respectively and a frame rate about 10 Hz. We also introduce a hierarchical generalized regression neural networks (H-GRNN) scheme to map eye and mirror parameters to gaze, achieving a gaze estimation accuracy of 92% under head movements. The use of H-GRNN also eliminates the need for personal calibration for new subjects since H-GRNN can generalize. Preliminary experiments show our system is accurate and robust in gaze tracking under large head movements.