The Research And Implementation Of High Frame Rate Real-time Eye-tracking System

Scientific research shows that 80% to 90% of information obtained by a normal person is from the eyes.It can be said that eye is the main organ for people to obtain outside information.Continuous research on eye tracking has led to the rapid development of eye tracking technology.Eye tracking technology has been successfully applied to many fields and it is believed that this technology will be applied to more fields in the future.With the increasing maturity of eye tracking technology,various types of eye tracking systems have emerged,such as head-mounted systems and desktop systems.This article introduces the research background and research significance of eye tracking technology at the beginning,then describes the types and principles of the current eye tracking technology and the research status of the domestic and foreign eye tracking systems,and then further introduces related theoretical basis of the desktop eye tracking technology.This eye tracking system is mainly composed of three parts: hardware,software and algorithm.The hardware consists of several infrared light source,an infrared camera,a functional board,and a transmission board.The software part can be divided into two layers: video acquisition layer and multi-threaded application layer.The algorithm part mainly includes image preprocessing,pupil location,C,P point solution and sight line point calculation.The implementation of the three parts of the system is described in detail in Chapter 4.Based on the realization of all the functional modules of the system,this paper tests the stability,real-time and accuracy of the eye tracking system based on the 3D eyeball model.The test proves that the system can work stably in real time,and the accuracy of the system is about 0.82°.The general work flow of the desktop eye tracking system is the acquisition of bright dark pupil images,bright dark pupil images preprocessing,using bright and dark pupil images to locate the pupil,getting the center of the pupil and the center of the spot,the solution of C,P points,and the solution of the sight line landing points.Using bright pupil images to locate the pupil is a critical step in the eye tracking system and can directly affect the real-time and accuracy of the eye tracking system.In order to improve the performance of the system,based on the advantages and disadvantages of existing pupil location algorithms,a pupil location algorithm based on image processing and machine learning is proposed in this article.The algorithm is proved to adapt to many situations such as wearing glasses and not wearing glasses.At the same time,the algorithm only needs the pupil image to locate the pupil position,so the algorithm indirectly improves the frame-rate of the eye tracking system.At the same time,the accuracy of the system is also correspondingly improved.The workflow of the algorithm is image filtering denoising,image interpolation reduction,sliding window,window histogram segmentation,candidate region center selection and precise positioning of the pupil.Through the test of six testers,the results show that the average accuracy of the algorithm is 97.8%,the average total time is 13.3ms,and the theoretical average frame rate per second can reach about 75 fps.The actual test shows that the algorithm has high real-time performance and robustness in most cases.At last,this paper implements a practical application of an eye movement control dome camera based on the eye tracking system.This application uses gaze point data to control the rotation of the dome camera.This article describes the implementation of this application in detail.At the same time,the algorithm part of this application is improved.The improved algorithm makes the dome camera run more stable with Gaussian smoothing fixation data and PID control.