Detecting Student Engagement in Classrooms for Intelligent Tutoring Systems

Detecting and tracking student engagement in a large classroom can help tutors or automatic learning systems easily control or summarize the situation. To come up with the advanced technique in machine learning, especially deep learning, nowadays, many schools can build an efficient system for supporting teachers or tutoring systems. In this paper, we propose a transfer learning method applying to a small dataset to classify student actions in the classroom. Another contribution is building a lightweight dataset with a limited number of images for each category for classification work. The experiments show the acceptable result in action recognition with high accuracy compared to other researches.