Cutting Edge Weapon Detection in Real-Time CCTV Videos

In today's world, ensuring public safety through advanced surveillance systems is critical. This paper presents a cutting-edge weapon detection system that operates in real-time using CCTV video streams, leveraging deep learning algorithms such as YOLO (You Only Look Once) and Detectron2. The system is designed to detect and classify weapons, such as guns and knives, with high accuracy while maintaining low latency. Through the use of optimized deep learning models, the system processes live video feeds and identifies potential threats in real-time, providing immediate alerts to security personnel. The proposed system is scalable, supporting multiple video feeds. The paper explores the methodologies employed to achieve real-time processing, including frame extraction, model inference, and post-processing techniques like Non-Maximum Suppression. Optimization techniques such as quantization, pruning, and hardware acceleration are applied to ensure the system runs efficiently on edge devices. Extensive testing was conducted in real-world environments, demonstrating the system's ability to handle diverse lighting conditions, occlusions, and varying weapon types. The results show significant improvements in detection accuracy, response time, and system reliability, making it a valuable addition to modern security solutions. Furthermore, the system adheres to privacy laws, ensuring compliance in surveillance applications. This paper contributes to the advancement of automated surveillance systems by providing an efficient and effective solution for weapon detection in public and private spaces.