A Hybrid Deep Learning Framework for Multi-Camera Object Tracking and Anomaly Detection in Surveillance

Modern surveillance systems face increasing challenges in real-time object tracking, behavioral analysis, and anomaly detection across distributed camera networks. Traditional single-camera solutions fail to address spatial discontinuity and occlusion issues in multi-camera environments. To overcome these challenges, this paper proposes a hybrid deep learning framework integrating Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) with attention-based data fusion for multi-camera object tracking and anomaly detection in large-scale surveillance networks. The proposed framework employs a CNN-based feature extractor to identify and encode object characteristics across multiple camera streams, followed by a spatio-temporal attention module that aligns overlapping regions and correlates object trajectories between cameras. The RNN-LSTM network models temporal dependencies to track object motion patterns, while an autoencoder-based anomaly detector identifies abnormal behaviors such as unattended baggage, unauthorized entry, or erratic movement. A multi-view consistency algorithm ensures smooth cross-camera transition tracking. Experimental evaluation on PETS 2009, TownCentre, and Avenue datasets demonstrates the framework’s superiority, achieving a tracking accuracy (MOTA) of 94.7%, an IDF1 score of 92.5%, and an anomaly detection accuracy of 96.2%, outperforming existing CNN-LSTM and transformer-based baselines. The proposed hybrid framework enhances situational awareness, robustness, and scalability, making it suitable for deployment in smart cities, public safety monitoring, and industrial surveillance.