FireModelAI: A Smart Framework for Fire Detection, Localization, and Classification Indoors

This study presents FireModelAI, a context-aware fire monitoring system tailored for dynamic indoor environments. The proposed architecture employs a dual-stage detection strategy, initiating with a low-power Passive Infrared (PIR) sensor for initial fire detection, followed by a cloud-based deep learning module using Detectron2 for bounding-box-based object detection and triangulation-based spatial localization. The system estimates fire coordinates with a mean absolute error of 1.26 cm. It classifies fire types—Class A (solids), Class B (flammable liquids/gases), and Class C (electrical)—based on the presence of nearby object types, such as electronic devices or combustible furniture, to select appropriate extinguishing agents and reduce unintended damage. The proposed system achieved a detection accuracy of 88.46%, confirming its potential for precise and adaptive fire response i n real-world indoor scenarios.