Occlusion Robust Cognitive Engagement Detection in Real-World Classroom

Area of Science:

• Educational Technology
• Computer Vision
• Artificial Intelligence

Background:

• Cognitive engagement is crucial for learning and can be inferred from observable behaviors.
• Automated detection of cognitive engagement offers valuable pedagogical insights.
• Challenges in automated detection include object occlusion, class similarity, and variance within classes.

Purpose of the Study:

• To propose an effective object detection model for automatically measuring cognitive engagement.
• To address the challenges of occlusion, inter-class similarity, and intra-class variance in engagement detection.

Main Methods:

• Development of the Object-Enhanced-You Only Look Once version 8 nano (OE-YOLOv8n) model.
• Integration of an improved Inner Minimum Point Distance Intersection over Union (IMPDIoU) Loss within the YOLOv8n framework.
• Creation and utilization of a real-world Students' Cognitive Engagement (SCE) dataset for evaluation.

Main Results:

• The proposed OE-YOLOv8n model demonstrates superior performance in detecting cognitive engagement.
• The model achieved a precision of 92.5% across five distinct engagement classes.
• Experiments on the custom SCE dataset validated the model's effectiveness.

Conclusions:

• The OE-YOLOv8n model offers a robust solution for automated cognitive engagement detection.
• The IMPDIoU Loss effectively enhances detection accuracy in challenging scenarios.
• This approach has the potential to provide instructors with real-time insights into student engagement.