Real-time visual detection of early manmade fire

The detection of early manmade fire carries profound meaning in warning systems to prevent fire-related terrorist attacks. Despite a large number of work on fire detection in the computer vision literature, there is no specific method for early stage manmade fire detection, our best knowledge. Compared to traditional fire detection, there is less information on early stage man-made fire for detection, because of the small scale and short appearance. In this paper, an algorithm is proposed by combining the characteristics of fire with human activity. Experimental comparisons have shown that our algorithm has outperformed the state-of-the-art fire detection methods, as the initial combustion is considered.