Research On Skeleton-based Teacher-student Behavior Pattern

With the development of education informatization in the big data age,how to use modern intelligent technology to automatically recognize behaviors of teachers and students in the classroom,and accurately measure the student engagement is the urgent problem to be solved.It is benefit to comprehensively analyze the quality of teaching and learning efficiency.Most of existing researches on teaching and learning behavior analysis only focused on verbal behaviorsand ignored the importance of non-verbal behaviors in the class.And they paid sole attention on student behaviors when measure the student engagement.However,student learning behaviors are often closely related to teacher behaviors.The same action of student under different guidance of teacher actions may represent different engagement.So it is not objective and accurate enough to measure engagement if ignoring the connection between students’ and teachers’ behaviors.Aiming at those problems,this paper propose the posture-based Teacher-Student Behavioral Engagement Pattern(TSBEP)to measure student engagement more accurately by both teacher and student actions.According to the correlation between teacher and student actions,it divides the teacher-student behaviors into 19 kinds of action pattern pairs,and defines the student behavioral engagementthey represent.In addition,a big video dataset of teacher-student actions in traditional classrooms is established for action recognition.And the Spatial Temporal Graph Convolutional Networks(ST-GCN)based on transfer learning is used to recognize actions of teaching and learning in videos,which achieves a higher accuracy than other image-based deep learning models.Finally,the effectiveness and objectivity of TSBEP are verified from the theory and practice by statistical analysis of teacher-student behaviors in real traditional classes videos.