Eye Gaze Direction Estimation Based On Images

Eye gaze direction estimation provides the direction information of the eye, and it has great prospects of application in the field of intelligent human-machine interaction, true three-dimensional video presentation and so on. The eye gaze direction estimation based on images contains a number of key technologies such as face detection and location, eye detection and gaze direction analysis, in addition to using the camera, there is no need for additional hardware, and it can realize the natural and non-intrusive detection, so the development of simple, practical eye gaze direction estimation method has great value of application.Based on the analysis of facial features, an eye gaze direction estimation scheme based on image is proposed. In the scheme, after locating the orbit, it estimates the eye gaze direction by calculating the distance between the center of the eyeball and the inner side of the orbit. First, by analysing the facial features, it concludes that the human eye in the fixed region of the face, that's to say, the position ratio is relatively fixed, and roughly confirm the eye region according to these values, and then calculates the inner position of the orbit. Then, two-dimensional modeles of human eye face with a variety of postures are built. According to the characteristics of parallel binocular motion, human eye models with different gaze directions are built, which is based on the face of straight, the face with the left side on, and the face with the right side on. By analysing these models, the regularity of the distance between the eye center and the inner side of the eye region was concluded. Finally, an algorithm is designed to estimate the gaze direction by calculating the difference of the distances of the left and right eye centers to the inner position of the orbit of each side.In the laboratory environment, the gaze direction detection experiment is carried out on faces with a variety of postures. For the case of small angle tilt of the faces, thereal-time detection accuracy rate is higher than 80%. On the factors affecting the accuracy of detection, disrupting chemicals(such as glasses etc.) is the greatest factor, the face of inclination is the second. When an inclination of large angle occurs, the accuracy rate has a sharp decline. The main reason is that the orbital localization in this project is a method of roughly locating, which can be optimised to improve the accuracy of eye gaze direction estimation.