Gaze aided human computer interface

Abstract

ABSTRACTDuring the last ten years, researchers have experimented with using the eyes to control the computer. But many either use more than one camera or use a pan-and-tilt base, or both. Such methods add to system cost and restrict the systema??s dynamic properties. In this paper, a non-contact corneal-pupil reflection scheme using only one fixed camera to track the eye gaze is presented. A small manual focus lens is used in a camera without a pan-and-tilt base. A connected component labeling algorithm is designed to detect the pupils. After the calibration process, the head movement is tracked by a Kalman filter with a constant speed model. And the four eye corners are extracted and tracked by using the Lucas-Kanade method. Then the head rotation is compensated by two methods. First, the pan and roll degrees of the head are calculated by using geometric information. These degrees are used to compensate the measured view vectors. Second, a Radial Basis Function neural network (RBF NN) is trained with head pan. The features are the four eye corners and the view vector. Canny edge detector and Susan corner detector are used combined to find the eye corner precisely. Two applications are designed to test the system, one is the famous Quake game, and another is a Windows calculator.