> 연구성과 > 국외저널
사용후기
책임교수 |
이의철 |
논문명 |
3D Eye-Tracking Method Using HD Face Model of Kinect v2 |
논문종류 |
LNCS / LNAI / LNEE |
제1저자 |
김병철 |
교신저자 |
이의철 |
공동저자 |
- |
Impact Factor |
- |
개제학술지명 |
Lecture Notes in Electrical Engineering |
Keyword |
Gaze Tracking |
게재일 |
2016 년 05 월 |
Despite the extensive research into 3D eye-tracking
methods, such methods remain dependent on many additional factors such as the processing
time, pose, illumination, image resolution, and calibration procedure. In this
paper, we propose a 3D eye-tracking method using the HD face model of Kinect
v2. Because the proposed method uses accurate 3D ocular feature positions and a
3D human eye scheme, it can track an eye gaze position more accurately and
promptly than previous methods. In an image captured using a Kinect v2, the two
eye-corner points of one eye are obtained using the device’s high-definition
face model. The 3D rotational center of
the eyeball is estimated based on these two eye-corner points. After the center
of the iris is obtained, the 3D gaze vector that passes through the rotational
center and the center of the iris is defined. Finally, the intersection point between
the 3D gaze vector and the actual display plane is calculated and transformed into
pixel coordinates as the gaze position. Angle kappa, which is the gap between
the actual gaze vector and the pupillary vector, is compensated through a user-dependent
calibration. Experiment results show that the gaze estimation error was an
average of 47 pixels from the desired position.