Improved Fire Detection Using Mean-Shift-based Mode Detection

초록

Camera-based fire detection is an important application in intelligent surveillance system. In this paper, Mean-Shift-based mode detection is proposed to improve the performance of fire detection. Firstly, candidate fire pixels are detected based on the motion information. After thatThen, Mean-Shift is applied to cluster fire regions from candidate fire pixels. Furthermore, fire-color rules are defined to improve refine the clustering results. Finally, the flame flicker characteristic is used to classify the fire clusters from fire-like clusters. Experimental results with different situations and image resolutions prove demonstrate that the proposed algorithm can decrease the false detection and the miss rate significantly, especially in the situations with random wind or small fire.