Computer Vision-based Method to Detect Fire with Color Variation in Temporal Domain

Abstract

Because of false detection rate, many fire detection algorithm cannot easily become a part of alarm monitoring system, so in this paper, we proposed to detect fire precisely with color variation in temporal domain to reduce false detection rate. By capturing special characteristic of fire in temporal domain, we can get reliable experimental result tested by 24 video sequences. With our algorithm, we can detect fire in more than 90% of cases tested by fire and non-fire 24 sequences, and also makes false detection less than 10%. And most of false detection are shown in unusual videos that have fire becoming extinguished and dancing people with moving camera.