Publication

Abstract : Real-time detection of student inattentiveness remains a significant challenge in online education, where conventional methods often fail to provide instructors with immediate or actionable feedback. The primary objective of this study is to address this gap by developing and evaluating the Student Attentive Geometry Group (SAGG) Detection System. This lightweight deep learning framework utilises facial orientation and gaze tracking to assess student attentiveness from a webcam stream. The system is trained and evaluated on a real-time dataset comprising 750 facial images depicting five distinct head poses (up, down, left, right, and straight). Employing three convolutional neural networks—VGG16, VGG19, and a proposed custom lightweight CNN (CL-CNN)—the CL-CNN model achieved a balanced performance with 93.2% accuracy and inference times of 20 ms per frame on the CPU and 5 ms per frame on the GPU. Continuous attention tracking allows the system to notify instructors when distraction persists for more than 20 consecutive frames. The SAGG Detection System offers a scalable, non-intrusive method to enhance virtual classroom engagement, laying the groundwork for personalised and adaptive online education.