Abnormal Behavior Detection of Examinees in Surveillance Images Based on Deep Learning

This paper introduces a deep learning model framework tailored for identifying abnormal behavior among candidates in monitoring images. The model seamlessly incorporates three pivotal techniques: The Spatiotemporal Graph Convolutional Network (ST-GCN) effectively captures both the spatial and temporal characteristics of examine-takers' movements. Gated cycle units (GRUs) enhance sequence understanding to ensure sensitive capture of persistent or transient abnormal behavior; The attention mechanism optimizes information screening and focuses on key dynamics that affect the detection results. First, spatio-temporal map is constructed by skeleton extraction, and ST-GCN extracts spatio-temporal features on this basis. Then, GRU is used to capture the long-term dependencies between sequences and maintain the continuous analysis of behavioral sequences. Furthermore, the added attention mechanism is dynamically weighted, highlighting the temporal and spatial characteristics directly related to abnormal behavior. Finally, through the full connection layer and Softmax function, the model outputs the classification probability of abnormal behavior. The framework effectively integrates multi-level feature processing and information screening. Experiments show that the model performs well in a variety of abnormal behavior recognition tasks, and provides strong technical support for intelligent invigilation system, which is expected to play an important role in the automation of educational invigilation in the future.