Study And Implementation On The Algorithm Of Eye Tracking System

Eye tracking technology is considered as one of the most effective means in the research of visual information processing. Its development makes the experimental conditions tend to be more natural and improves the bionomics validity largely. Also its research results have been widely applied to many fields, such as psychology, work efficiency study, computer science and kinematics .In this paper, we first discuss the development process and related theories of the eye tracking technology. Then, one kind of block diagram is proposed, which represents a realizing eye tracking system. It consists of eye image processing subsystem, calibration subsystem, optical subsystem and data analysis recorder subsystem. However, we focus on the first two sub-systems in this paper.In the eye image processing subsystem, we first transform eye images to binary-image with adaptive threshold, in order to locate the corneal reflection spots. Then we detect feature points of the edge of pupil by searching the large derivative of a radial in the eye image without the corneal reflection spots. Finally, we use the feature points to fit the edge of pupil and locate pupil center.Also, one more method is advanced to deal with the interruption problems of eye data caused by a blink, which is based on optimum linear prediction.In the calibration subsystem, the mapping relationship is determined by data fitting , between eye images and scene images. We choose a gaze point in a new eye image as the input of the data fitting function. Then we consider the result as a mapping point and mark it in a synchronized scenes image. Also, kinds of mapping function with different orders and different modes of eye tracking used by the author are to experiment with coupled eye data or no coupled eye data in order to find an optimal function. And it is proved that a second order mapping function is an optimal function for the calibration subsystem .The function is based on a pupil-cornea tracking mode and the cross-coupled relationship between eye image data.