Early indoor occluded fire detection based on firelight reflection characteristics

Abstract

Indoor fire detection is a challenging task and plays a key role in disaster management. The early stage of a fire is the best stage to extinguish the fire. However, early detection of indoor fires is difficult because the early stages of fires are easily occluded in complex indoor environments. Therefore, a method based on firelight reflection characteristics is proposed for early fire detection in occluded indoor environments. First, the characteristics of fires occluded by complex environments are described by analyzing the characteristics of firelight reflection. Second, a highly sensitive method for foreground recognition is developed through use of strategic background updates and a block binarization threshold, which are suitable for detecting the weak changes caused by occluded fires in videos. Finally, a multiexpert system is established for occluded fire detection by extracting the changing characteristics of the area in which the firelight reflection occurs, including spectral variability, motion persistence, and regional expansion. The accuracy and run time of the system are evaluated based on a large dataset to verify our method. Moreover, our proposed approach is discussed in detail in terms of effectiveness and applicability, and the results show that our method can be effectively applied in indoor occluded fire detection.