Yolo-Panic: A Real-Time Ai-Based Gesture Recognition System for Smart Panic Alarm System

Abstract

Rapid and effective responses during emergencies are critical in today’s fast-paced environments, including industrial facilities, commercial hubs, public institutions, and transportation networks. Traditional panic alarm systems largely depend on physical hardware—such as buttons or switches—which may be inaccessible or ineffective in high-stress or hazardous situations. To address these challenges, we introduce YOLO-Panic, a real-time AI-based gesture recognition system designed to activate smart panic alarms using intuitive hand gestures. Built upon a customized YOLO (You Only Look Once) deep learning architecture, the system is trained to detect a specific panic gesture—four fingers extended with the thumb folded inward—with exceptional accuracy.

The YOLO-Panic model achieves outstanding performance metrics, including a precision of 98.56%, mAP@50 of 99.30%, and recall of 97.99%, ensuring high reliability with minimal false positives. To enhance spatial gesture interpretation, a keypoint-based module is also integrated, achieving 94.06% mAP@50 and 83.27% mAP@50–95, enabling robust, fine-grained analysis of hand poses. The system operates with a high degree of efficacy in real-time, rendering it exceptionally suitable for application in dynamic and bustling environments.

As cities evolve into smarter and more connected ecosystems, YOLO-Panic offers a vital layer of safety and situational awareness. Its contactless, AI-driven approach aligns seamlessly with the goals of smart city infrastructure, where intelligent surveillance and rapid emergency response are paramount. By enabling intuitive human-machine interaction for crisis communication, YOLO-Panic represents a scalable, adaptive solution that enhances public safety, reduces emergency response times, and supports the development of resilient, technology-enabled urban environments.