Study On Gaze Tracking In Video Image Sequences

Eyes are important organs for perceiving the world, gaze direction stands for the point of interests. Gaze tracking technology gets the points of interest by detecting the gaze direction of human. As the progressive development of electronic technology, more and more research study on gaze tracking, some gaze tracking systems have been used in the field of human-computer interaction. Gaze tracking technology will be developed in the fields of network availability, advertising, packaging design and automotive engineering and so on. However, due to the limitation of existing eye tracking technologies, gaze tracking system always has shortcoming of low precision, limiting head movement and intrusion. To solve above problems, the technique of gaze tracking in video image sequences has been studied. The constraints on the user's head have been significantly relaxed. The accuracy and stability of the system have been improved.In this thesis, such achievements have been achieved:(1) Proposed the combined method of head pose estimation and two-dimension gaze tracking, using the head pose parameters correct the distance between facial feature points which used for gaze tracking, reduces the errors caused by head rotation.(2) The head pose estimation algorithms for gaze tracking is studied, a 3-dimentions head pose estimation method is proposed. In this algorithm, assumed that the head is a cylinder, rotation of head is seen as the rotation of this cylinder. Changing parameters continuously, make current face texture match the reference texture, to get the posture parameters of current image. Using the characteristic of little changes between adjacent video frames, the head texture of previous frame can be used to estimate the head posture of next frame, reduces the amount of computation and improved the computational accuracy.(3) The sub-pixel techniques of edge location algorithms are studied. In the process of gaze tracking, using sub-pixel technology to position facial features (Iris center and outer corner point); Ellipse fitting method is used in position the Iris center by sub-pixel Iris edge points. It reduces the gaze tracking error caused by features location error, the accuracy of the system has been improved.(4)Using a simple geometric model of head, collecting feature vector of certain gaze directions, computes gaze direction by the existing corrections between feature points and gaze direction. Which can accurately estimate gaze direction, enable the timely request. The system of gaze tracking in video image sequences was developed, verified the accuracy and stability of this system.