Real‐Time Student Abnormal Behavior Detection Based on Distributed Edge Collaboration and Multi‐View Fusion Under Classroom Scene

ABSTRACT Efficient detection of abnormal student classroom behavior is of great significance for improving teaching quality and maintaining classroom order. However, traditional methods based on centralized cloud computing face the challenge of poor real‐time performance. In addition, existing video‐based anomaly action detection networks are difficult to deploy effectively on resource‐efficient edge servers. Therefore, this article proposes an efficient variant spatiotemporal graph convolution (ST‐GCN) model based on distributed edge collaboration learning. First, we deploy multiple edge servers and cameras in the classroom to form a distributed collaborative perception network. Second, we integrate the multi‐scale graph convolution and the multi‐view fusion mechanism to enhance the robustness of feature expression. Finally, we construct a collaborative learning mechanism to improve the generalization ability of abnormal action recognition. The experimental results on our self‐built classroom abnormal behavior dataset show that our proposed method outperforms existing graph convolution networks in terms of detection accuracy.