Research And Application Of Gaze Estimation In Online Examination Monitoring

The application of computer-aided education in learning evaluation and the proposal of paperless examination have triggered profound changes in the form and content of evaluation,and the research on online examination system has also been carried out under this circumstance.After the outbreak of the new crown epidemic in 2020,most courses and examinations are changed to online formats,which promotes the development of online examinations.Like offline exams,there are also various cheating behaviors in online exams,and the flexibility of the examination location increases the difficulty of invigilation,so how to ensure the fairness and impartiality of online exams is an important research topic.Aiming at the shortcomings of the existing online examination monitoring system in the market,this thesis applies the techniques of gaze estimation and head pose estimation in computer vision to online examination monitoring to detect cheating behavior only due to the change of line of sight.The results of gaze estimation can be used to predict line of sight direction and fixation target,which has a good application prospect in medicine and psychology.The movement of the eyeball is the main reason for the change of sight.In addition,head posture also has an impact on sight.Therefore,this thesis uses the head posture to assist the sight judgment.In general,attention can be judged by head posture,and head posture estimation is beginning to be used in tired driving,intelligence classes and medical research.The main research contents of this thesis are as follows:(1)Aiming at the problem that some detailed features appearing in the deep network are filtered,a head pose estimation model based on improved ResNet is proposed.The backbone of the model uses the improved ResNet50 network.The improvement points include the fusion of deep and shallow features and the addition of SE attention mechanism.The improved backbone network can effectively utilize deep and shallow features,and can effectively solve the loss caused by the difference between the deep feature channels,the network model can reduce the error of head pose estimation.(2)In order to better extract eye features without reducing the spatial resolution,the dilated convolutional neural network is applied to the gaze estimation.The network model uses PreActResNet8 as the first half of the trunk network to avoid insufficient features extracted caused by too few convolutional layers,thus reducing the model preformance.In the latter part,the dilated convolutional neural network is used to increase the receptive field without reducing the spatial resolution,thus improving the accuracy of model prediction.(3)Design and implement an online exam monitoring system based on gaze estimation,including front-end interface design,database design,and server-side design.The monitoring modules involved include front-end interface monitoring,server-side authentication,and gaze estimation.Finally,the debugging of the system is completed,and the system can normal operation.