Development of a Multimodal Fire Detection Framework

Abstract

This study proposes a multimodal fire detection framework utilizing multiple sensors to overcome the limitations of existing single-sensor-based fire detection systems, which often struggle with early detection and high false alarm rates. The proposed framework consists of data collection, edge computing, and cloud analysis layer, which operate independently and exchange real-time data. The data collection layer integrates and collects multi-sensor data. The edge computing layer preprocesses the collected data in real time, performs analysis such as image analysis and sensor anomaly detection, and independently responds to network failures with local alerts. The cloud analysis layer integrates and manages data transmitted from the edge and applies an advanced deep learning model that fuses image and sensor features to determine the final fire risk. Developing a system based on this multimodal fire detection framework is expected to contribute to minimizing loss of life and property by reducing false alarms and increasing the accuracy of early fire detection through information complementation.