DEEP LEARNING-BASED CLASSROOM ANOMALY DETECTION USING OBJECT-CENTRIC TEMPORAL MODELING

Abstract: Detecting the unusual patterns in video footage plays a crucial role in ensuring safety by alerting authorities to potential risks. In many real-world scenarios, delayed identification of abnormal events such as fights or suspicious behavior can lead to serious consequences. This project focuses on developing an automated video anomaly detection system using deep learning and computer vision techniques. The system processes video footage by extracting frames and detecting foreground objects using the YOLOv8 algorithm. Relevant spatial features are extracted using a ResNet-based convolutional neural network, and temporal patterns are learned using a Bidirectional Long Short-Term Memory network. The extracted features are analyzed to identify abnormal activities using statistical methods such as Z-Score-based anomaly scoring. When an abnormal event is detected, the system generates real-time alerts by displaying warning messages, triggering an alarm sound, and sending an email notification to the user. The proposed system is implemented using Python and integrated with a Streamlit-based web interface for visualization. This approach improves real-time montoring, reduces response time, and enhances the effectiveness of surveillance systems.

Keywords: Video Anomaly Detection, Deep Learning, YOLOv8, BiLSTM, Temporal Analysis