Research And Implementation Of Desktop Gaze Tracking System Based On Differential Light Exposure Model

The gaze tracking system estimates the point of gaze by collecting relevant data from the users' eyes when they observe the outside world,providing a more effective way for humancomputer interaction of the disabled.However,the current low-cost desktop gaze tracking system is usually not suitable for human-computer interaction due to its bad precision,and the high-precision desktop gaze tracking system is much expensive because most of them use high-end global shutter image sensors;and most gaze tracking systems can only be applied to a specific PC operating system,which is not conducive to the gaze tracking algorithm research and equipment popularization.In this paper,we combined bright and dark pupil differential exposure pupil detection technology,which is originally applied to high-precision eye movement equipment,with low-cost rolling shutter image sensor,and develop a low-cost cross-platform desktop gaze tracking system based on 3D eyeball model.The problem of using differential exposure technology on low-cost rolling shutter image sensor is that the exposure characteristics of the rolling shutter lead to interference in the process of differential exposure.In this paper,a rolling shutter differential exposure model is proposed and deduced.This model can be used to calculate the exposure start time and exposure time length of differential exposure for image sensors with different parameters,so to achieve the goal of using cheap rolling shutter image sensors to construct highprecision gaze tracking system.In addition,this paper also proposes solutions of the bright and dark pupil image frame losing problem,the system cross-platform problem and fixtion detection problem in the gaze mouse application.In this paper,we firstly describe the general design of the desktop gaze tracking system,including FPGA video capture board,USB3.0 data transmission board and the PC software which contained frame layer and algorithm layer in the second chapter.Then,in the third chapter a differential lighting exposure model is designed for the rolling shutter image sensor.It is used to calculate the light exposure parameters of the bright pupil light and dark pupil light.The counting compensation algorithm and frame combine technology are proposed to solve frame loss problems.In the fourth chapter,to improve image data transmission capacity and device compatibility,A UVC firmware is implemented on CYUSB3014,and technologies such as QT,Direct Show and V4L2 are used to solve the cross-platform problem of the host software.In the last part of the fourth chapter,an application used for disabled human-computer interaction,the gaze mouse,is designed and implemented based on our desktop gaze tracking system,and the I-DT point of gaze classification algorithm used in the application is improved.In the process of research and implementation of desktop gaze tracking system,the major contribution of this paper are as follows:(1)A differential light exposure model for rolling shutter is proposed(2)The counting compensation algorithm and frame combine technology are proposed to solve frame loss problems when images are captured(3)Design the FPGA image capture board and program it(4)Design and implement a UVC firmware(5)Develop a cross-platform host software framework(6)Develop a gaze mouse application for disabled human-computer interaction and improved I-DT algorithms.In the end,this paper implement a low-cost,cross-platform and high-precision desktop gaze tracking system which can capture two 1280x720 monochrome images synchronously with 60 fps,and use bright and dark pupil differential pupil detection technology and 3D eyeball model to estimate point of gaze.The system's mean error is 0.8 °.