Real-time fire detection system based on dynamic time warping of multichannel sensor networks

Abstract

Fires are considered one of the most devastating accidents that could occur within a building. However, existing fire detection systems result in late fire detection or produce a high false alarm rate due to the complex patterns and temporal dynamics found in sensor signals that come from the various types of fires. This paper develops a novel fire detection algorithm that is based on multiple sensors in different locations to provide reliable real-time fire monitoring. The proposed fire detection algorithm takes into account both current and past sensor measurements and evaluates the similarity of sensor signals based on a dynamic time warping distance measure. Moreover, the proposed algorithm adaptively selects the sensors that are critical in detecting a fire in an early stage based on the developed k-out-of-P fire voting rule, which effectively combines decision rules from P multichannel sensor signals. A case study was performed using a real-life fire dataset obtained from the National Institute of Standards and Technology to demonstrate the effectiveness of the proposed approach.