Enhancing Intelligent Surveillance
M. Aimola Davies Anne, M. Brindha, N. Sivakumaran · Advances in computational intelligence and robotics book series · 2025
This chapter presents a hybrid deep learning framework for intelligent surveillance, integrating anomaly detection, violence recognition, object tracking, and person re-identification to enhance real-time threat detection. The system employs Variational Autoencoders (VAEs) and Long Short-Term Memory (LSTM) networks for anomaly detection, while ResNet50 and 3D Convolutional Neural Networks (3D CNNs) extract spatial and temporal features for violence recognition. YOLO and DeepSORT enable real-time object detection and tracking, ensuring continuous monitoring in complex environments. OpenPose and LSTM networks refine pose estimation for behavioral analysis, while deep metric learning enhances person re-identification across multiple camera views. Extensive evaluations on benchmark datasets validate the system's accuracy and robustness. This chapter details the methodologies, implementation, and validation strategies, offering insights into the role of deep learning in strengthening security applications for real-world surveillance.