Deep Learning-Based Anomaly Detection and Object Classification for Campus Surveillance
Shakshi Richhariya, David Wang, Jerry Zeyu Gao · 2025
Manual CCTV review on large campuses is impractical. This paper surveys and proposes a deep learning-based pipeline for anomaly detection in campus video surveillance that combines two stages: frame-based event detection and object classification. Our method uses a CNN-LSTM model to predict normal video sequences, flagging frame-level anomalies when significant deviations occur. To add semantic context and actionable insights, we integrate a YOLO object detection module, which classifies detected objects and checks rule-based criteria such as object type, count, and zone. For example, the system can both detect a scene-level anomaly and classify it as “unauthorized vehicle” or “crowd” in a restricted area. We introduce and benchmark our approach using a real campus video dataset. Our solution improves accuracy and interpretability, delivering clear, actionable alerts suitable for real-world deployment.