AUTOMATED RECOGNITION OF UNUSUAL HUMAN BEHAVIOR IN CCTV VIDEOS USING NEURAL NETWORKS

Abstract

Suspicious human activity recognition has become a critical component of modern intelligent video surveillance systems, driven by the rapid advancement of deep learning and the increasing demand for automated security monitoring. Existing research demonstrates significant progress in activity understanding, anomaly detection, and spatiotemporal feature learning using models such as 3D-CNNs, ConvLSTMs, sparse reconstruction, and Generative Adversarial Networks (GANs) [11–22]. Recent surveys highlight the growing adoption of AI-powered surveillance in public safety, smart cities, and predictive policing applications [2–10, 23–24]. Building upon these developments, this work proposes an enhanced suspicious activity recognition framework that integrates robust person detection, deep spatiotemporal feature extraction, and anomaly classification to accurately identify irregular behaviors in real time. Leveraging insights from sparse coding–based methods, fully convolutional anomaly detectors, and temporal regularity learning, the system aims to improve detection accuracy while reducing false alarms. The study contributes a unified analysis of traditional and contemporary deep learning approaches, emphasizing their strengths, limitations, and applicability to real-world surveillance scenarios. The proposed model aligns with emerging trends in AI-driven security and offers a scalable solution suitable for large-scale urban deployments