Fire video recognition based on flame and smoke characteristics

Abstract

The fire detection methods by using pure flame or pure smoke often lead to the phenomenon of missing alarm. This paper presents a novel fire video recognition method based on both flame and smoke. Firstly, fire regions of interest are detected using Kalman Filter. Then, three major features of flame including flickering, spatio-temporal consistency and texture feature based on Local Binary Pattern (LBP) are extracted from flame-like regions. Three major features of smoke including flutter feature, energy analysis and color feature are extracted from smoke-like regions. Finally, D-S evidence theory fuses two evidences generated by Neural Network to recognize fire images. Experimental results show that the proposed method can significantly reduce missing alarm rate and false alarm rate.