ARTDet: A Novel Attention-Based Deep Learning Network for Real-Time Head Posture Detection of Students

Head posture detection of students in classroom environments is crucial for monitoring engagement and optimizing the learning experience. However, this task remains challenging, especially in crowded and dynamic classroom settings where small targets are prevalent. To address this, we propose ARTDet (Attention-based Real-Time Detection for Classroom), a novel network for the detection of the head posture of students based on deep learning. To boost detection performance in classroom environments and address multi-student scenarios, we incorporate the NAM (Normalization-based Attention Module) into the backbone, greatly refining the model's capacity to detect small targets in challenging backgrounds. In addition, we create a custom classroom dataset, manually annotating each student with bounding boxes and posture classifications, providing valuable resources for training and evaluation. Extensive experimental validation demonstrates that ARTDet outperforms existing methods in both quantitative and qualitative assessments.