Real-Time CCTV Weapon Detection for Public Safety: Design, Implementation, and Responsible Deployment

Public spaces such as schools, shopping malls, workplaces, and transport terminals remain vulnerable to weapon-related violence. Conventional CCTV systems depend on human operators whose vigilance degrades during prolonged monitoring, leading to missed detections and delayed responses. This study designs, implements, and evaluates a real-time AI-enhanced CCTV pipeline that automatically detects visible weapons — pistols, rifles, and knives — and issues alerts to security personnel within sub-second latency. A YOLO-based deep learning architecture was trained on 600 labeled images and evaluated across 50 CCTV-style video scenarios covering indoor, outdoor, low-light, and occluded conditions. The system achieved a mean detection accuracy of 92.4% (precision 90.1%, recall 93.8%, F1-score 91.9%), an average alert latency of 480 ms, and a false-negative rate of 3.7%. The paper further proposes ethical guidelines for deployment grounded in the EU AI Act, the IEEE Ethically Aligned Design framework, and UNESCO's Recommendation on the Ethics of AI.