The Eye Tracking System Algorithm Based On Video Image

Eye movement system is a key technology in the field of human-computer interaction. It has a wide range of application in security, vehicles, assistive, instrument, military, medical, computer science, and many other fields. The eye movement, is refers to the movement of scanning, staring and blinking of the pupil. This paper studies the key algorithm of eye movement system based on video image.By analyzing the geometric model, this paper builds a simple platform for collecting the eye movement video, which includes one CCD camera and a computer. The main work is as follows:1. This paper used three kinds of preprocessing method on eye movement video image, which laid a good foundation for the follow-up work. Then comparing the advantages and disadvantages of several kinds of face detection algorithm, this paper introduces YCb Cr color model for face detection and the Gabor transform method for detecting the eye location.2. Sobel operator is applied to the face image for the edge detection. In order to highlight horizontal edge in the eyebrow area, we use the improved Sobel operator to deal with face image from the eye movement video. We find the two features in eyebrow area: horizontal projection values in eyebrow eye area nearby are relatively large, and it has a certain width. According the two features, we separate the eyebrow area from the face detection results. After that, for the purpose of tracking the monocular eye movement process, this paper proposed the segmentation of eyebrow area using the vertical gray-level projection. Meanwhile, horizontal gray-level projection is used to determine the monocular eye bar area among the segmented image of eyebrow area. Finally, we use the simple search algorithm to determine the trace point which is the accurate results of the positioning pupil center. We choose the 15 frames as a result of simulation experiment through the whole process of eye movement.3. The innovation point in this paper is that we proposed a kind of visual tracking method based on spatio-temporal context after accurate location of the pupil. We choose the accurate positioning results of pupil as a condition of track initialization. We tracked the whole process of glancing to the left and right, the up and down, and the irregular eye movement. We drew the path of eye movement through the glancing among the MATLAB. After that, we measured the accuracy of the tracking by comparing the number of abnormal points. This algorithm can effectively overcome the impact of slight head rotation and blink of eye through the tracking process. Furthermore, we discovered that it can reach the speed at 0.014 seconds per frames by processing 10 eye movement video, which is a relatively fast processing speed that could meet the requirement of real-time performance and robustness.