Computer vision based fire detection in color images

In this paper, a generic rule-based color model is proposed for fire pixel detection. The proposed algorithm exploits the YCbCr color space to separate the luminance from the chrominance for effectively addressing the issue of illumination variations. A set of rules defined on the Y, Cb and Cr color components together with the developed chrominance model on the Cb-Cr color plane are used to detect the fire pixels in color images. The performance evaluation of the developed color model and rules was conducted on a set of color images, which consists of fire, non-fire and fire-like regions. The proposed rule-based color model achieves up to 99% correct fire detection rate with 31.5% false-alarm rate. The results are compared with two other color models and show higher correct detection rate and lower false-alarm rate.