Research And Application Of Learning Attention Detection Method Combining Head Pose And Gaze Estimation

Driven by modern information technology,e-learning models have emerged.However,the gap between time and space makes communication between teachers and students insufficient,and students' learning efficiency is greatly affected by their self-control.Students are prone to attention deficit.Therefore,it is necessary to quantify and analyze the participation of students in e-learning to help students monitor their learning state and engage them deeply in the learning process.Head pose estimation and gaze tracking are important research fields in computer vision and artificial intelligence.In recent years,more and more researchers have paid attention and interest.In this paper,we utilize the characteristics of head pose estimation and gaze tracking based on deep learning algorithms and build an efficient system for students' attention recognition in e-learning environments.It effectively improves the recognition accuracy rate and enabling students to obtain a more intelligent learning experience.The work of this paper can be summarized as followed:(1)Based on detailed analysis and research on head pose estimation tasks,this paper proposed a novel algorithm,namely,anisotropic angle distribution model.It is based on two key observations which can be briefly generalized as following:1)For a certain pose,variations of the head pose in yaw and pitch within the same interval are different;2)In yaw angle direction,variations in the interval of 0° to 30° and 60° to 90°are smaller than variations in the interval of 30° to 60°.Furthermore,this model can be learned in an end-to-end manner via a constructed convolutional neural network.Experimental results on the public datasets indicate that the proposed algorithm achieves state-of-the-art performance.(2)A new gaze tracking method based on the multi-loss convolutional neural network is proposed and implemented.The multi-loss network has two independent fully connected layers,which predict the yaw angle and the pitch angle of the gaze respectively.Each loss function of the angle is combined with classification and regression.Experiments on public datasets show that the proposed gaze tracking method has better accuracy and robustness than other methods.(3)A learning state recognition system based on the combination of head pose and gaze tracking for students in an e-learning environment is designed.Experiments show that the system can identify and analyze students' learning attention objectively and accurately.