Real-Time Student Attentiveness Monitoring System using YOLOv5

Abstract

This paper presents real-time classroom monitoring system designed to detect and quantify student attentiveness using deep learning and computer vision. By leveraging YOLOv5 for behavior recognition (reading, writing, hand-raising, and focused gaze)  and integrating head pose estimation, the system provided objective, dynamic engagement metrics. The model was trained using the SCB-Dataset merged with a custom focus dataset, achieving a precision of 1.00 and mAP@0.5 of 0.681 on validation data. With real-time video stream processing, mobile webcam integration, and a responsive dashboard, ClassCam supported educators in fostering engagement through data-driven decisions.