Multi-feature based fire detection in video

This paper presents an automatic system for early fire detection in video. In many applications, it is desired to increase the detection sensitivity and, more importantly, improve the reliability of the detection system through improving the false alarm immunity. The reliability is significantly improved such that fire detection systems can automatically control the fire suppression system. The use of multi-feature based detection technology can offer the most promising means for achieving both the improved sensitivity to real fires and the reduced susceptibility to false alarms. In this paper, we present a new mathematical tool based on fractal dimensions, along with the chromatic features, to make a raw localization of fire regions first. Further, we consider the dynamic features of the early fire in video and show how they can be applied to fire detection to improve the fire detection performance. The experiments performed on a large number of scenes demonstrate the effectiveness of the proposed method.