Improvement of gaze estimation robustness using pupil knowledge

This paper presents an eye gaze estimation system which robust against various users. Our method utilizes an IR camera mounted on glass to allow user's movement. Pupil knowledge such as shape, size, location, and motion are used. This knowledge works based on the knowledge priority. Pupil appearance such as size, color, and shape are used as the first priority. When this step fails, then pupil is estimated based on its location as second priority. When all steps fail, then we estimate pupil based on its motion as the last priority. The aim of this proposed method is to make the system compatible for various user as well as to overcome problem associated with illumination changes and user movement. The proposed system is tested using several users with various race as well as nationality and the experiment result are compared to the well-known adaptive threshold method and template matching method. The proposed method shows good performance, robustness, accuracy and stability against illumination changes without any prior calibration.