Research on behavior recognition algorithms in classroom scenarios

The quality of teaching has always been a primary concern in the field of education. With the introduction of the smart classroom concept and advancements in hardware and software technologies, there is a need for an automatic and unbiased tool to evaluate teaching quality. Based on the principle of computer vision, this paper designs a behavior recognition network to classify students' behaviors in class. These classification results can provide data basis for evaluating the concentration of students and the attractiveness of teachers in the classroom. Also this information is essential for teachers to analyze their courses, optimize course rhythm, and improve teaching quality. Based on the data set of the university's classroom, this paper designs the whole process of student behavior identification in the classroom. This paper introduces a TSM module in the network according to the actual needs of real-life use, and also makes lightweight improvements to the network so that it can be better adapted to real-life classroom student behavior recognition tasks. After improvements, the behavior recognition network used in this paper achieved 88.98% accuracy. This result is optimal compared to other common behavior recognition networks. After lightweight measures, we reduced the size of the model by one ninth of the original model, and the recognition speed is only half of the original model. This greatly improves the practicability of the algorithm. Therefore, the methods proposed in this paper are valuable and have potential application prospects.