Eye Tracking Algorithms Based On Head-mounted Eye Tracker

Human-computer interaction technology is one of the hotspots of contemporary computer technology research.With the diversified development of the material world,people's demand for multi-channel human-computer interaction methods is becoming more and more urgent.The two-way nature of visual interaction makes it more intuitive and natural in the human-computer interaction process than the mouse and keyboard.So,it is of great significance to study the gaze tracking technology.The traditional desktop gaze tracking system is not suitable for users' large-scale motion scenes.The existing head-mounted gaze tracking technology in China has problems such as poor universality and large complexity.Therefore,this topic makes further research on the line-of-sight tracking algorithm based on the head-mounted eye tracker with reference to the foreign advanced line-of-sight tracking equipment,which reduces the complexity and improves the stability of the system.In this study,a method of pupil center mapping is proposed for the headmounted line-of-sight tracking device to estimate the fixation point.The method fully utilizes the feature that the position between eye image capturing camera and the human eye is relatively fixed,the origin point of the coordinate system in the eye image is selected as the stationary reference point,extraction of the corneal reflection spot and the corner position of the eye is avoided,which simplifies the algorithm processing flow of the pupil-corneal reflection method and the pupilcorner method.In the part of eye tracking algorithm research,this study proposes a method of secondary extraction of pupil center,and completes the research design of the eye tracking combination algorithm.The pupil region segmentation image is obtained by image preprocessing,then edge detection and boundary fitting method is used to obtain the pupil center point.This method avoids the misdetection of eyebrows and eyelids as pupil.In the process of establishing the line-of-sight tracking model,three kinds of gaze point mapping models based on polynomial fitting are assumed to meet the requirement of users with high degree of freedom,and the gaze point calibration method is explored.In order to verify the eye tracking accuracy and gaze tracking accuracy of the algorithm,a series of experiments are also designed.The results show that the proposed method based on secondary extraction of pupil center and Log edge detection and RANSAC-based boundary fitting method can accurately extract the pupil center point.The accuracy of eye tracking is basically close to the existing research level.The experimental results show that the average processing speed of the line-of-sight tracking algorithm proposed in this paper is more than 35 frames per second,which satisfies the real-time requirements of the line-of-sight tracking system in human-computer interaction.The final result shows that the accuracy of line-of-sight tracking is related to the distance of the human eye to the gaze point plane.The closer the distance is,the smaller the error is.When the distance between the human eye and the gaze point plane is about 720 mm,the tracking error of the gaze point is 1.02°,corresponding to 12.8 mm in the real space.It satisfies the basic requirements for the distance between human and machine and tracking accuracy in human-computer interaction.In conclusion,the line-of-sight tracking algorithm based on pupil center mapping proposed in this paper achieves accurate tracking of eye movement state and human eye fixation point,and meets basic human-computer interaction requirements.