Research On Gaze Estimation Method Of Head-mounted Eye Tracking System

The gaze estimation is a new intelligent human-computer visual interaction method,which is a technology to calculate the gaze direction or gaze position of the eye by constructing a model between the eye features in the eye image and the point of regard.Based on the analysis of the integrated structural characteristic of the headmounted eye tracking system,this thesis focuses on two parts,including the algorithm of the pupil feature extraction under near-infrared light,and the establishment of a gaze estimation model based on this feature.The main work and research results of this thesis are as follows:(1)This thesis describes the current research status and trends of gaze estimation technology at home and abroad,and derives the research background and significance of the subject based on the characteristics of the head-mounted eye tracking system.It specifically analyzes the algorithm of eye feature extraction,2D and 3D gaze estimation models,and compares and summarizes the current advantages and disadvantages of the gaze estimation methods.(2)Aiming at the head-mounted eye tracking system where the human eye is close to the camera and the near-infrared light source illuminates the human eye to obtain a dark pupil image with obvious contrast,this thesis improves a algorithm of the pupil feature extraction under near-infrared light.In the types of complex environment data sets,the average pupil detection rate under 5 pixel errors is 79.10%,and under 10 pixel errors is 85.00%,and the average extraction rate is 11.63ms/frame.In the application experiment,the algorithm of pupil detection under the near-infrared light source illumination in different lighting environments,different experimental users and different directions has promising results.(3)Focusing on three performances requirement of head-mounted system,including the accuracy of estimated angle error,the complexity of eye tracking system,and the accessibility of users,this thesis improves a simplified 3D geometric gaze estimation model based on pupil contour,and compares the accuracy of the gaze estimation angle error.The analysis show that the average angle error in this thesis is0.53°,and the standard deviation angle error is 0.24°;in terms of the complexity of system,the minimum configuration of the system is a single camera and a single light source;in terms of the accessibility of users,the user can estimate the gaze point of regard through the single-point calibration process of free head movement.These meet the actual application requirements of human-computer interaction.Research on the gaze estimation method of the head-mounted eye tracking system,which is not only has the academic value of studying human psychology and behavior,but also has a wide range of practical research significance applied to scenarios such as virtual reality and augmented reality.