Abstract: An effective, four-stage fire-detection algorithm used to automatically detect fire in video images was presented in this paper. An adaptive Gaussian mixture model was used to detect moving regions in a video clip. A fuzzy C-means (FCM) algorithm was adopted to segment the candidate fire regions (fire and fire-colored objects) from these moving regions based on the color of fire. Some special parameters were extracted based on the tempo-spatial characteristics of fire regions; these parameters included the area randomness, surface roughness and motion estimation of fire. Finally, these parameters extracted from the third stage were used as input feature vectors to train a support vector machine(SVM) classifier, which was then used by the fire alarm to distinguish between fire and non-fire. Experimental results indicate that the proposed method outperforms other fire detection algorithms, providing high reliability and a low false alarm rate.