AeroGuard: intelligent UAV framework for fire and disaster management

Abstract

This research proposes a multimodal system for early fire detection and disaster mapping using a UAV (quad-copter). The system combines fire/no-fire/human in fire image classification, sound analysis, and sensor data within an UAV framework. Initially, surveillance cameras detect early fire in-dicators such as luminous smoke and textured flames. If fire is confirmed, an alarm triggers the quad-copter, which captures aerial visuals and identifies human presence by analyzing human voice in fire prone area. The system then maps the affected area and marks safe zones to aid in rescue operations. A custom fire, disaster, and sound(primary) dataset was developed. The fire detection model achieved an accuracy of 80.32%, while the sound detection model reached an accuracy of 97.14%. The system also significantly reduced false fire detection and drone noise cancellation. The four modules are proposed to integrate into an android application to enable real-time emergency response and rescue planning in this framework.