YOLO-Based Physical Education Classroom Behavior Recognition and Teaching Analysis

The physical education classrooms offer a difficult situation of video analysis as they have small and distant targets, many occlusions, swift changes in behavior, and intense student-teacher relationships. The solution to these problems is a YOLO-inspired approach to the recognition of classroom behavior and teaching analysis in physical education classrooms presented in this paper. The framework combines detecting targets and multi-object tracking with temporal behavior discrimination and classroom-statistics generation into one pipeline. To enhance robustness in crowded and fast changing settings, a lightweight attention model, multi scale feature enhancement, class balanced optimization and trajectory level temporal modeling are incorporated. Moreover, selective augmentation can be used to reduce the influence of the sample imbalance on the minority classes, particularly teacher-student interaction and transition movement. The experimental findings indicate that the proposed method performs better than its representative baseline counterparts in terms of detection accuracy, identity association consistency, and behavior recognition with efficient inference. The system also offers readily interpretable teaching indicators such as activity density, participation rate, waiting time ratio, and interaction intensity, which will help in evaluating classrooms based on data and enhancing the quality of instruction.