Eye-Gaze Estimation by Using Features Irrespective of Face Direction

摘要

Most of the previous methods for constructing a human machine interface that uses a line of sight are divided into two groups according to the method of line detection. One uses a special light source or a hardware device, and the other uses software for image processing. Our proposed method, which is of the latter type, determines whether a person gazes at a camera lens or not by using only a single facial image captured by a video camera. In previous methods of this kind, it is common understanding that the directional angle of a head with respect to the world coordinate system and that of the eye in relation to the head are estimated independently, and the direction of the line is determined by adding one to the other. A feature of the proposed method is that eye-gaze estimation is conducted by using a pair of marks that are attached at a special position on a facial surface, without considering the head direction in space. Our proposed method can be realized easily, and can estimate the eye-gaze in a short time. Experimental results show the effectiveness of the proposed method.