Abstract: As conventional fire detection technology has many limitations, this paper proposed a fire recognition method based on image. Firstly, the image color space was transformed from RGI3 to YCbCr and a coordinate system was set up then the Cb and Cr values of the fire sample pixels were depicted. Secondly, an ellipse was drawn up to ring up the most of the points so as to create an elliptic equation and a two-dimensional normal distribution function which value out of the ellipse was zero. Lastly L-M algorithm was used to optimize the parameters of normal distribution function. So the recognition of the fire was converted to judge if the normal distribution function value of a given couple of Cb and Cr isgreater than zero. I}he method of this paper has good real-time and recognition performance.

Key words: Fire recognition, YCbCr color space, Normal distribution, L-M algorithm