A Deep Learning Model for Suspicious Event Detection in CCTV Footage

Automated detection of suspicious activities in CCTV footage remains a critical challenge for modern surveillance systems. While deep learning approaches have shown promise, existing methods often struggle with real-world complexities such as occlusions, varying lighting conditions, and adversarial vulnerabilities. This paper presents a novel 3D-ResNet-18 architecture enhanced with attention mechanisms and multi-scale feature fusion (AMF) to improve anomaly detection in surveillance videos. The proposed model effectively captures spatio-temporal patterns while focusing on semantically relevant regions, enabling robust performance across diverse scenarios. Evaluated on benchmark datasets (UCF-Crime and ShanghaiTech), our approach achieves superior accuracy (90.4%) and computational efficiency (45 FPS), outperforming state-of-the-art methods. Additionally, we address privacy concerns through ethical preprocessing and evaluate the model’s resilience against adversarial attacks. The results demonstrate significant improvements in both detection performance and real-world applicability, making the proposed framework a viable solution for intelligent surveillance systems.