3D Facial Landmarks Video Analytics for Student Engagement

The rapid transition to online and hybrid classrooms has increased the need for automated systems that are capable of interpreting student engagement through non-verbal cues. Traditional techniques rely on two-dimensional (2D) video analytics or the use of specialized hardware such as eye-tracking glasses. Both of these approaches limit the scalability and accuracy of the results. In this work, we introduce a 3D facial landmarkbased video analytics framework that quantifies student engagement without the need for specialized equipment. Using MediaPipe for real-time 3D landmark extraction and the Close Eye Aspect Ratio (CEAR) model for engagement classification, the proposed work distinguishes attentive from drowsy states in students during recorded sessions. Experimental results showed that 3D CEAR measurements provide more stable and accurate engagement indicators compared to 2D baselines. These findings validate the potential of 3D vision analytics to enable a more accessible and hardware-independent engagement assessment method.