Student in-class behaviors detection and analysis system based on CBAM-YOLOv5

Analyzing the students’ in-class behaviors is able to intuitively reflect their learning states, which are important for teaching evaluations and for designing instruction. In order to automatically detect and recognize students’ behaviors through surveillance video, we design a CBAM-YOLOv5 detection and analysis system. By fusing Convolutional Block Attention Module (CBAM), CBAM-YOLOv5 could eliminate the effect of background information and effectively extract robust feature of student behaviors. A new dataset including six in-class behaviors, such as drinking, study, trance, writing, hand_up and using_phone, is built to training and validating the proposed CBAM-YOLOv5. Experiments demonstrate that the effectiveness and robust feature extracting capability of the proposed CBAMYOLOv5. The detected students' in-class behaviors are classified and summarized, and the teaching quality report is timely fed back through Wechat applet.