Abstract: Flame Detection based on videos is important for research on fire prevention. I}o improve the detection per- formance,GMM(Uaussian Mixture Model) which can adapt the change of background was used to detect moving sub- jects at first. Then the features of color and changing area were extracted from these moving objects. Finely, these fea- tures were applied to recognize whether there is a flame or not in the scene. The method can not only effectively detect flame in videos, but also save computing time in non-fire scene.

Key words: Flame detection, GMM, Color model, Area changing