Intelligent Safety-Redundant IoT-Edge System for Indoor Gas Leakage Detection

Abstract

This paper presents the development of an intelligent gas leak detection system designed to enhance safety in residential household environments. The system integrates an MQ-2 gas sensor, a Raspberry Pi V4 (24 GFLOPs), with Camera Module, and AHT20 sensor to monitor gas concentrations, detect smoke, and track temperature conditions. Real-time alerts are triggered through local buzzers and instant messaging notifications when hazardous conditions are identified. Image object detection model, trained on a dataset of fire and smoke images, aids in detecting smoke and fire in the vicinity. Five test-cases were designed to simulate real-world household conditions in order to test the safety-redundancy of the decision system in isolation. In 71% of the simulated real-world household conditions the alert response is issued correctly; with precision at 0.68 for the sensor-based gas leak decision system. At 1.1 GFLOPs the SSD MobileNetV2 FPNLite model (preliminarily) fine-trained on the FASD [7] dataset reached 43.6% mAP50; a suitable time complexity level for on-device or on-microcontroller operation. Performance is compared to the YOLOv8 model (8.7 GFLOPs). The preliminary experimental results demonstrated the system's utility to detect indicators for indoor flammable gases, fire and smoke hazards. Further field testing is necessary to assess its performance. This implementation provides an affordable and scalable solution for gas leak detection, offering added safety-redundancy.