3D Eye Model-Based Gaze Tracking System with a Consumer Depth Camera

Most existing gaze tracking systems are high-cost, intrusive and difficult to calibrate, and some rely on the infrared illuminant. However, such systems may not work outdoor and meet real-time requirements. This paper proposes a non-intrusive system based on the 3D eyeball model, which does not need the exact infrared illuminant and complicated calibration process and allows the natural movement of the head. In the proposed system, Kinect is used to track the iris center and face model of the person, and the 3D information is easy to obtain. At the same time, point cloud registration algorithm is applied based on feature points in the face model sequence to obtain accurate head pose estimation results. In this paper, a personal calibration process is also proposed to obtain the gaze model parameters for different users, such as the eyeball center and angle kappa. The proposed method has good adaptability to the change of illuminant and head movement. In the actual operating environment, the system speed reaches 30 fps, which can meet the requirements of real-time control.