A Fast Image-Based Fire Flame Detection Method Using Color Analysis

Abstract

ABSTRACT Fires usually cause serious disasters. Thus, fire detection has been an important issue to protect human life and property. In this paper, we propose a fast and practical real-time image-based fire flame detection method based on color analysis. We first build a fire flame color feature model based on the HSI color space by analyzing 70 training flame images. Then, based on the above fire flame color features model, regions with fire-like colors are roughly separated from each frame of the test videos. Besides segmenting fire flame regions, background objects with similar fire colors or caused by color shift resulted from the reflection of fire flames are also extracted from the image during the above color separation process. To remove these spurious fire-like regions, the image difference method and the invented color masking technique are applied. Finally, the fire flame burning degree is estimated so that users could be informed with a proper fire warning alarm. The proposed method was tested with twelve diverse fire flame videos on a Pentium 1.73 GHz processor, 512 MB RAM at the processing speed of 30 frames per second. The experimental results were quite encouraging. The proposed method could achieve more than 96
detection rate on the test videos on an average. In addition, the system could also correctly recognize fire flames within one second on the initial combustion from the test videos, which seems very promising.