Research Of Attention Analysis System Based On Gaze Estimation Neural Networks

With the spread of educational anxiety and the increasingly fierce competition for high-quality educational resources,education has shown a trend of fast-paced.Teachers be distracted to observe students’ learning status and obtain feedback information to improve teaching methods during lectures,which brings a lot of burden to teachers and it’s not conducive to the overall progress of students.Therefore,some auxiliary means are urgently needed to help teachers judge the students’ learning status,especially students’ attention status.In order to solve the problem,the work of this article uses the3 D gaze estimation neural network based on the full-face image to estimate the direction of the student’s gaze,then calculates the location of the student’s gaze point in the teaching environment,finally quantify attention based on the relevance of the gaze point to the lecture,the main work is as follows:（1）Design and optimize 3D gaze estimation neural networks TFGaze Net and SSRGaze Net.Aiming at the low-resolution characteristics of human eye images,the high-resolution neural network is used,and Transformer Encoder and soft stagewise regression are used in feature fusion.After optimization and testing,TFGaze Net has better performance,and has good generalization performance and stability for gaze estimation under different faces and various head poses.Compared with the baseline gaze estimation networks,TFGaze Net effectively reduces the number of parameters of the model at the cost of a small loss of accuracy.（2）An attention quantification analysis method based on gaze vector projection is designed.Using the binocular camera to obtain face images and depth information,the gaze estimation neural network estimates the direction of the gaze,combined with the classroom environment information,calculates the position of the student’s gaze point in the classroom,and then quantifies the student’s attention.It is not limited by the camera installation method and classroom environment,can adapt to unfamiliar scenes.（3）Deploy edge computing for attention analysis algorithms and gaze estimation neural networks,and conduct attention analysis experiments.The related algorithms and models were deployed on Atlas 200 DK,and the classroom teaching attention analysis experiment was designed in a simulated scenario.The experiment’s results showed that the algorithm was effective and feasible.