Learning behavior monitoring through drowsiness and screen-distance recognition

Abstract Real-time monitoring of students’ learning states is essential for improving engagement assessment and personalized intervention in online education. However, existing approaches often rely on computationally intensive facial expression analysis, gaze estimation, or multimodal behavior recognition, limiting their deployment in resource-constrained learning environments. To address these challenges, this study proposes a lightweight learning behavior monitoring framework based on drowsiness detection and screen-distance estimation. An improved object detection model, termed YOLO26-P2-CBAM, is developed by integrating a high-resolution P2 detection head, the Convolutional Block Attention Module (CBAM), and the Wise-IoU (WIoU) loss function into the YOLO26 architecture. The proposed model enhances small-target perception and improves robustness under complex learning conditions. Based on this model, a browser–server collaborative monitoring system is implemented to perform real-time yawning detection and eye-to-screen distance estimation using webcam video streams. In addition, parallel processing and batch inference strategies are introduced to improve computational efficiency in multi-user scenarios. Experiments were conducted on the public SCB-dataset and two self-constructed datasets for yawning detection and screen-distance estimation. The proposed model achieved an [email protected] of 63.2% on the SCB-dataset, outperforming the baseline YOLO26 by 4.2% points, while also achieving the highest Precision (62.2%) and F1-score (60.9%) among all evaluated methods, indicating a better balance between detection accuracy and target coverage. On the self-constructed yawning dataset, it achieved an [email protected] of 86.8%, while screen-distance estimation yielded a mean absolute error of 2.12 cm. The average inference time was approximately 12.5 ms per frame, satisfying real-time application requirements. These results demonstrate that the proposed framework provides an effective and deployable solution for intelligent learning-state monitoring in online education.