Research On Key Technologies Of Infrared Eye Movement Image Processing For 3D Eye Tracker

Eye tracker is an instrument used to collect eye movement information.Combining the collected eye movement information with the stimulus content,it can reflect the subjects’ psychological and brain activities in a lateral way.It can provide a basis for research on some social behaviors and even brain diseases.Compared with the twodimensional eye tracker,the stereoscopic eye tracker can provide the subjects with more immersive stereoscopic stimulus content.so it is expected to obtain more accurate three-dimensional eye movement information.However,the type of eye movement information collected is mostly infrared images,and the effect of general image processing methods on eye movement analysis is inapplicable.In view of this problem,this paper made the following research:1)In this paper,an optimization algorithm S-RITnct based on deep learning was proposed to improve the accuracy of infrared eye image segmentation.In the design of S-RITnet algorithm,the optimization of fuzzy infrared eye image quality was the core.Based on real-time eye image semantic segmentation network RITnet,the selfestablished eye image data set was adopted to accurately semantic segmentation of sclera,iris and pupil.Finally,the mIoU value obtained by S-RITnct in the verification set reached 0.9406.F1 score was 0.972,and the segmentation effect of S-RITnet was the closest to that of labeled image.The results showed that S-RITnet successfully improved the accuracy of infrared eye image segmentation.2)In order to solve the problem of line of sight calculation,based on GazeML algorithm,the input improvement and the realization of 3D fixation point calculation and visualization window were carried out.The optimized GazeML algorithm can calculate the infrared eye movement image(video)source,and generate visual fixation point image(video)combined with the stimulus content(video).3)3D fixation point calibration platform was designed for 3D eye tracker(S3D-ET)developed by our research group.In this work,a set of 3D calibration parameters dedicated to infrared eye movement image analysis of the eye tracker was proposed.The GazeML algorithm was integrated to analyze and optimize the eye movement to complete the design of 3D fixation point calibration platform.Based on the S3D-ET instrument,47 people were recruited to test the validity of the 3D calibration platform.They completed the eye movement gaze task respectively,and completed the comparative analysis of the accuracy of the eye movement algorithm before and after the calibration through the 3D calibration platform.The results showed that the 3D calibration parameters of the platform can be effectively used to calibrate 3D fixation points in S3D-ET,and the platform has a positive impact on the improvement of GazeML’s application accuracy in infrared eye movement image analysis.The research on the key technologies of infrared eye movement image processing under the 3D eye tracker is tentative.With the continuous development and deepening of this research,the self-developed 3D eye tracker is expected to further expand the comprehensive research in the fields of social behavior,brain neurology and other fields.