An early detection algorithm for small flame based on vision system with variant field of view

A fast detection method for small flame area in variable visual fields is proposed to solve the difficulties of flame identification caused by the small flame area, unapparent features and its easy-interferential in fire prime. A Pan-Tilt-Zoom is used to track and center the small suspected fire fields in a large range region. Then, the zoom lens is handled to enlarge the small area, and to enhance the 'multi-resolution' visual recognition feature. Then, an adaptive region growing algorithm is adopted to detect and segment the suspected flame region. Finally, a confidence probability model is built and the suspected probability is used as the quantitative parameters to confirm flame recognition. Experimental results show that the flame imaging quality is improved clearly, and that the small fire detection precision is greatly increased. The system can be widely used in early fire finding and long-range flame detection and warning.