Live Streaming Surveillance Video Upscaling Using Generative AI With Anomaly Identification
Deepika Amol Ajalkar, Abhishek Mahindrakar, Akshay Popale, Ashish Meshram, Prabal Khillarkar · 2024
This research addresses the challenge of low-resolution CCTV footage in criminal investigations through the application of generative artificial intelligence (AI). Leveraging Generative Adversarial Networks (GANs) and Super-Resolution Convolutional Neural Networks (SRCNNs), our approach focuses on real-time upscaling to improve the quality of surveillance video. The proposed system involves preprocessing low-resolution frames and utilizing generative AI models to produce high-resolution images, enhancing crucial visual details for effective crime scene analysis. Extensive experiments on real-world CCTV datasets demonstrate the superiority of our method in terms of accuracy, visual fidelity, and real-time processing compared to conventional upscaling techniques. Furthermore, our research explores the novel use of generative AI for crime scene identification. By automatically analyzing the enhanced footage, the system detects and identifies potential evidence such as objects, vehicles, and individuals. This capability contributes to more efficient criminal investigations, aiding law enforcement agencies in solving cases effectively. While advancing the field of computer vision and AI-assisted law enforcement, our work provides a practical and effective solution for real-time CCTV footage upscaling and crime scene identification. The deployment of this technology has the potential to significantly improve the speed and accuracy of investigations, thereby enhancing public safety and security. However, ethical considerations, including privacy and potential biases, are crucial aspects that require careful attention during implementation.