3-D Gaze Tracking with a Monocular Vision System

• Main Authors: Chien-Yu Chen, 陳建宇
• Other Authors: Sheng-Wen Shih
• Format: Others
• Language: zh-TW
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/02301226766741895050

碩士 === 國立暨南國際大學 === 資訊工程學系 === 94 === Gaze tracking is an important and interesting topic in the domain of computer vision and man-machine interaction. Through tracking user's gaze direction, it is possible to improve the efficiency of accessing a computer. Therefore, a gaze tracking system can not only provides another way for human-computer interaction, but also expands the applicability of computers. The line of sight (LoS) is a 3-D line, and thus a 3-D vision technique is needed to estimate the LoS accurately. The most convenient way to implement a 3-D gaze tracking system is to use stereo vision. However, a stereo vision system will suffer from the problem of dealing with the complicated refraction computation of the eye glasses. In this thesis, we proposed a 3-D gaze tracking method which needs only a single camera. The proposed method can achieve non-contact and non-intrusive gaze tracking while allowing user's natural head movement. This monocular 3-D gaze tracking method requires only one calibrated camera and two near-infrared LEDs. The gaze information is inferred by using the corneal reflection of the LEDs and the refracted pupil boundary. The gaze tracking error caused by the measurement noise is evaluated with a simulation program. Improvements are suggested with the simulation results to suppress the tracking error to an acceptable level. Through extensive computer simulations, we will show the gaze tracking accuracy under both systematic error and the measurement noise.