FIREVISION: An Intelligent Smoke and Fire Detection System Using Computer Vision with the BLIP Model and IoT Connectivity
This project proposes the development of an intelligent smoke and fire detection system using computer vision integrated with the Internet of Things (IoT). The solution aims to improve the efficiency and reliability of traditional fire detection methods, reducing false alarms and enabling faster and more accurate responses. The system is implemented in Python and integrates several libraries, including OpenCV for image capture and processing, the BLIP model for automatic caption generation and visual content analysis, as well as Flask and Blynk Cloud for real-time remote monitoring and control. The architecture was designed in a modular fashi
Resource information
- Authors
- Silva Soares, Gian Luca, Augusto moura da Silva, Luciano, Matheus, Souza, Rhaissa Julliani, Souza, Vitor Amadeu
- Research area
- Artificial intelligence
- Research topics
- Computer Vision, IoT, Fire Detection, Automation, Artificial Intelligence
- Publication date
- 2026
- DOI
- 10.5281/zenodo.19769236
- Upstream last modified
- Aug 24, 2026
- License
- Creative Commons Attribution 4.0 International
- Last record update
- Nov 6, 2026
About this resource
This project proposes the development of an intelligent smoke and fire detection system using computer vision integrated with the Internet of Things (IoT). The solution aims to improve the efficiency and reliability of traditional fire detection methods, reducing false alarms and enabling faster and more accurate responses. The system is implemented in Python and integrates several libraries, including OpenCV for image capture and processing, the BLIP model for automatic caption generation and visual content analysis, as well as Flask and Blynk Cloud for real-time remote monitoring and control. The architecture was designed in a modular fashion, combining machine learning, automation, and IoT connectivity. Tests confirmed the system’s operation: it was capable of detecting flames with precision and sending instant alerts via e-mail, web interface, and mobile application. To evaluate model performance, an experiment was conducted with 50 images, resulting in an accuracy of 98% and an average inference time of 1.169 seconds per image, demonstrating low latency and near real-time operation capability. These results reinforce the viability of an accessible, scalable, and low-cost solution with potential for application in industrial, commercial, and residential environments.