This paper presents a fire detection algorithm using some relationship among the three colors representing fire flames in a sequence of video frames. The proposed method designed in two stages. In the first stage, the candidate regions with fire flame colors are extracted using a simple statistically driven color model. In the second stage, the actual fire regions among the extracted candidate fire flame-colored regions are identified by dropping those candidate fire flame regions due to the appearance of some fire-like color objects. For this, the optical flow in the candidate fire flame regions are considered. Experimental results show around 95% of accurate results.