Modification of YOLOv9 with Bi-Level Routing Attention for Student Behavior Monitoring in Crowded Classrooms

Monitoring student behavior in classrooms is a crucial aspect to create an effective and interactive learning environment. Computer vision technology, particularly through the YOLO (You Only Look Once) model, offers a solution for real-time student behavior detection. However, limitations in detecting small objects, such as students at the back of the class or partially obscured by other objects in crowded classrooms, remain a challenge. This research modifies the YOLOv9 model by integrating a Bi-Level Routing Attention (BRA) module into the backbone to enhance detection accuracy. Experimental results show that the modified YOLOv9+BRA model increases the detection accuracy of small objects, with an mAP50 score improvement up to 80.2% on the YOLOv9-c variant. However, this increase in accuracy is accompanied by a decrease in inference speed, indicating a trade-off between accuracy and real-time performance. This study contributes to the development of computer vision-based monitoring technology for dynamic and crowded classroom environments.