Breakthrough in Real-time Fire Detection for IoT Devices

Research from Posts and Telecommunications Institute of Technology presents a novel real-time fire detection framework, optimized for resource-constrained edge devices, which achieves high mean average precision for fire and smoke detection. The framework, utilizing the fine-tuned YOLOv10 model and Coral Accelerator, demonstrates exceptional performance and is suitable for low-cost real-time applications. The system's lightweight design ensures reliable deployment in remote areas with limited computational resources, maintaining high accuracy while minimizing false alarms.

Key Takeaways:

• The proposed framework achieves a mean average precision (mAP) of over 84% for fire detection and a maximum mAP50 of over 91%.
• The framework reduces inference time by 58% compared to CPU-based implementations, with a latency of just 1.7 seconds per frame.
• The system's lightweight design ensures reliable deployment in remote areas with limited computational resources and unstable network connectivity.
• Fine-tuning the YOLOv10 model components in conjunction with hardware acceleration ensures both prediction accuracy and improved inference response performance.
• Comprehensive evaluations confirm the system's robustness, scalability, and practicality under various operating conditions.
• The framework demonstrates exceptional performance in fire, smoke, and distracting object detection, with 84% and 91% mAP metrics, respectively.
• The Coral Accelerator module ensures a significant reduction in inference time, making the framework suitable for low-cost real-time applications.
• Posts and Telecommunications Institute of Technology researchers Trong Thua Huynh, De Thu Huynh, Du Thang Phu, and Anh Hao Nguyen, were involved in the research.
• The research contributes to the construction of a diverse dataset encompassing fire, smoke, and distracting objects, an element often overlooked in existing fire detection datasets.

Statistics:

• Mean average precision (mAP) for fire detection: 84.1%
• Maximum mAP50 for fire detection: 91.2%
• Reduction in inference time: 58%
• Latency per frame: 1.7 seconds
• Average resource usage: 50% (compared to CPU-based implementations)

Sources:

• VerticalNews
• Journal of Computing and Information Technology (cite: A Lightweight Real-time Fire Detection Framework for IoT Devices Utilizing Fine-tuned YOLOv10 and Accelerator Module. Journal of Computing and Information Technology, 2025,33(3):157-181)
• Coral Accelerator module
• Posts and Telecommunications Institute of Technology
• University of Zagreb Faculty of Electrical Engineering and Computing