Fire Detection in Radiant Energy Domain for Video Surveillance

This paper presents a novel approach of video fire detection by the features of fire extracted in radiant energy domain. Firstly, a fire color model in YCbCr color space is applied to extract fire-colored pixels as candidate regions of fire. Secondly, we convert the color space of the candidate regions into radiant energy domain through camera calibration in advance and model six features of fire with spectral irradiances to better present the physical characteristics of fire. Finally, a two-class SVM classifier with a RBF kernel is adopted to recognize fire from the candidate regions. A series of experiments have been carried out on two different datasets. Experimental results illustrate that our approach performs well when compared with other state-of-the-art methods.