Integrating ESP32-Based Edge AI for Real-Time Fire Detection and Emergency Exit Monitoring

¹ Faculty of Electronics Engineering 1 & SEMIT Lab, Posts and Telecommunications Institute of Technology, Vietnam
² Faculty of Electronics Engineering 1 & EDA Lab, Posts and Telecommunications Institute of Technology, Vietnam
³ VNU International School, Vietnam National University, Vietnam

^* e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
^** e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
^*** e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
^**** e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
^† e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
^‡ e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

Industrial fire incidents pose serious risks to human safety and cause significant economic losses. In contrast, traditional fire detection systems often suffer from high latency, false alarms, and limited monitoring of emergency exit doors. To address these issues, this paper proposes a low-cost Edge AI system on an ESP32 microcontroller for real-time fire detection and emergency exit door state recognition without modifying existing infrastructure. The system integrates real-time image processing and environmental sensing using a lightweight MobileNetV1 model optimized for edge deployment, with all inference performed locally to ensure low latency and autonomous operation. Reliability is improved through temporal decision stabilization and multi-sensor consistency checking based on temperature and smoke data. Experimental results show 97.32% accuracy with 128 ms latency, while the fire class achieves 100% detection accuracy with no misclassification, demonstrating strong suitability for real-time industrial applications.