Study Of Head-Mounted Gaze Tracking Algorithm

Gaze tracking technology, as the latest kind of human computer interaction, obtains the eye images from camera and then estimates the locations of the gaze points from establishing the relationship between the gaze points and the eye features. The combination of gaze tracking technology and augmented reality can track user's line of sight in order to get the user's area of interest in the scene and can also reflect the characteristics of the combination of virtual reality and reality. Head-mounted gaze tracking system, which fixes the infrared light sources and the cameras around the human eyes though the helmet device, provides a simple, efficient and two-way interactive method for augmented reality system, so it has an important research significance and a broad application prospect. This thesis stated from the particularity of the structure of head-mounted gaze tracking system and then focused on the research of the eye movement feature extraction algorithm and the establishment of gaze mapping model. The major work and the research results of this thesis are as follows:1) The research of eye movement feature extraction under near-infrared light source. The human eyes are closer to the camera under head-mounted gaze tracking system. The infrared light sources and the fronted mirror lead to uneven contrast of the human eye images and make the deformation of the eye features. Considering above problems, this thesis selected the pupil and the iris as the eye movement features. According to the idea of getting the eye feature contour following by locating the eye feature center, this thesis proposed an automatic pupil threshold segmentation algorithm, which combined the OTSU algorithm and the histogram peak value searching threshold compensation algorithm, to achieve accurate segmentation of the pupil region. The iris region was also segmented by using the histogram iteration algorithm. Finally the pupil center and the iris center were extracted by using Sobel edge detection algorithm and RANSAC algorithm.2) The establishment of a 3D gaze mapping model based on head-mounted gaze tracking. Considering the precision of gaze estimation and the overall complexity of gaze tracking system, this thesis proposed a new 3D gaze mapping model based on pupil and iris, which used only one infrared light, one camera and four calibrate points on the screen. The three dimensional spatial model of pupil and iris was made by using the two dimensional center of the extracted eye movement feature. Then the gaze points were estimated by combining the 3D line of sight direction vector and the line of sight angle information.Simulation experiments and comparative analysis show that the feature extraction rate can be up to 25 frames per second and the pupil and iris feature extraction accuracy is 96% and 95%, respectively. So the eye feature extraction algorithms in this thesis keep a good balance between the extraction accuracy and the extraction rate. In addition, the accuracy of gaze estimation is. So the 3D gaze mapping model proposed in the thesis can ensure the gaze estimation accuracy of head-mounted gaze tracking system and greatly reduce the complexity of the whole structure and the calibration process of the system in the same time.