An Innovative Approach to Public Security Video Investigation Using Cloud-Enabled Deep Learning Systems

Abstract

This study introduces a Cloud-Enabled Deep Learning Framework aimed at enhancing public security through intelligent and automated analysis of surveillance video footage. Leveraging the power of deep learning and cloud computing, the proposed system is capable of accurately detecting complex human behaviors and identifying unusual or suspicious activities in real-time. The core of the framework employs a hybrid Convolutional Neural Network–Recurrent Neural Network (CNN-RNN) architecture, optimized and trained using the DCSASS dataset, which is specifically designed for security-based surveillance applications. The end-to-end system comprises video preprocessing, frame extraction, label generation, and deep learning-based classification integrated seamlessly within a scalable cloud infrastructure. This integration ensures not only improved computational efficiency but also better support for real-time data handling and large-scale video processing. Experimental results demonstrate the superior performance of the proposed model compared to a baseline ResNet-50 architecture. The hybrid model achieved an accuracy of 94.9%, precision of 94.2%, and recall of 94.4%, significantly outperforming the benchmark and indicating the robustness of the framework in classifying both normal and anomalous behavior in surveillance settings. By hosting the system on a cloud platform, the framework gains the advantages of scalability, flexibility, and faster response times, which are critical in real-world public safety applications. The study underscores the potential of reducing human monitoring efforts while increasing the efficiency and effectiveness of threat detection systems. Future work will explore expanding the dataset, incorporating multi-modal data such as audio or sensor inputs, and validating the system’s performance through real-time field deployment.