Teacher behavior-related object detection with an edge-enhanced and dynamic multi-scale network toward metaverse-oriented educational scenarios

Teacher behavior analysis is important for understanding instructional patterns and can provide perception support for metaverse-oriented educational environments. This paper focuses on teacher behavior-related object detection in classroom environments and discusses its potential role as a perception module for future metaverse-oriented educational systems. To address challenges such as scale variation, complex backgrounds, and inaccurate localization, an enhanced YOLOv8-based method is proposed. Specifically, an edge-enhanced feature extraction module is introduced to improve boundary representation, and a dynamic inception module is designed to capture multi-scale features. In addition, a Focaler-Shape-IoU loss is adopted to enhance bounding box regression accuracy. These improvements collectively strengthen detection performance. Experiments on a self-collected classroom dataset show that the proposed method achieves competitive results in terms of precision, recall, and mAP@50 compared with several mainstream detectors. The method can effectively handle complex classroom scenarios. Although this work focuses on object detection, it can serve as a foundation for further teacher behavior analysis and may provide foundational perception support for future metaverse-oriented educational applications.