Research On Calibration And Eye Movement Classification Algorithm Of Bracket Eye Movement Tracking Device

People's visual information occupies the vast majority of the proportion relative to the total content obtained from the outside.Only about 20% of the content is not from the eyes.Eyetracking devices which is used to record the characteristics of human eye trajectories while processing visual information are increasingly used and have great potential for development,yet suffer from disadvantages such as high prices and complex designs.Therefore,in this paper,we will combine the characteristics of bracketed vision tracking devices,overcome some limitations of the existing eye-tracking methods,and design a high-performance and low-cost bracketed eyetracking device.The main research work of this thesis includes the following contents:This article builds the software and hardware platform of the eye-tracking system.Users can use this system for calibration,eye tracking,and visualization of results.The hardware system includes the display part and the adjustment part.In order to collect clear and bright high-quality eye images,there are also main cameras and point light sources.In terms of the software system for eye tracking,the gaze point estimation algorithm is studied,as well as the analysis of the eyetracking data obtained by this algorithm to improve the accuracy of the eye movement type classification algorithm.An improved smooth tracking calibration method is proposed to overcome the limitations of poor usability of multi-point calibration operation and the complexity of equipment of the already available simplified calibration scheme.It can optimize the eye-tracking information collection process during the calibration process without increasing the equipment and obtain the best mapping point pairs of eye-tracking data and calibration points.Firstly,this paper proposes a rationalization method to remove noise and obtain valid data for eye-movement trajectory matching.Secondly,for the constraint that the user lags behind the calibration point,instead of using empirical values to set a fixed threshold for each user,the interrelationship number method is used to obtain the dynamic delay time in real-time for the user's attention detection.Finally,this paper explores the optimal parameter setting in the smooth tracking calibration method through experiments to ensure system accuracy and achieve the gaze point estimation.For the gaze point data obtained from gaze point estimation,a new offline hybrid eyemovement classification algorithm is proposed to analyze the data to obtain the information of eye-movement types with research significance.The algorithm performs two classifications for the three eye movement types.Firstly,saccades are extracted,and then fixations and smooth pursuits are distinguished.For each classification,a modified DBSCAN algorithm is used to initially segment the eye-movement tracking data before using the HMM algorithm to improve the classification accuracy.For the two parameters included in the algorithm,Eps and min Pts,adaptive thresholding based on kernel density estimation theory is used to finally achieve the reliable effect of automatically classifying eye movement types without the need for large amounts of training data and careful configuration of thresholds.