Classroom Attention Analysis System Based On Head Pose Estimation

Classroom attention analysis aims to capture rich semantic information to analyze students' responses to the classroom.With the development of information technology and artificial intelligence,more and more technologies are applied to classroom scenarios,providing more vitality to the concept of smart classrooms.However,there are some challenges in constructing a unified attention model in the classroom scene.First,the classroom scene has a large number of students,each student has a different face size,and the student's face resolution is relatively low.At the same time,each student is a separate individual and will have its own set of behaviors that are difficult to analyze.Aiming at these characteristics in the classroom scene,this article addresses the above problems from two aspects,the specific work is as follows:(1)Aiming at the small faces and low-resolution characteristics of students in classroom scenarios,this paper proposes a head pose estimation model in classroom scenarios.First,a face detector was constructed based on the network proposed by the Visual Geometry Group(VGG)of Oxford University.At the same time,a detection module was added to each scale branch of VGG to achieve multi-scale face detection.Then the features are combined by classification and regression to obtain more accurate head information.Finally,in training,the idea of data fusion are used to simultaneously train different quality pictures to make the model more robuster.Tested on public datasets WIDER FACE and AFLW,the experimental results show that our method has good performance.(2)In view of the difficulty of uniform analysis for different students,we constructed an attention analysis module.First,the attribute Euler angles common to each student are obtained through the head pose estimation model,and then the Euler angles are used as inputs,and the constraint relationship between the Euler angles is used to calculate.Finally,the angle between the seat and the blackboard is used to generate Spatial information to modify student Euler angles to further improve accuracy.Experimental results on classroom data set to demonstrate the performance and practicability of our method.Based on deep learning models,this paper aims at improving the performance of the head posture estimation module and the attention analysis module for the low-resolution features of the student's face and head and the difficulty of analyzing the students' attention in a classroom setting.The proposed method improves the detection and detection methods of students' attention in classroom scenes,realizes the improvement of traditional methods,and has certain reference value for target detection in classroom scenes and attention detection in natural scenes.