Fire Detection Using Spatial-temporal Analysis

Fire detection is an important issue of modern security sensing system. In this paper, we propose a vision-based fire detection algorithm. The proposed approach integrates color, spatial and motion information to locate fire regions in video frames. Potential fire regions are detected by modeling the fire color with Gaussian mixture model. Based on some characteristics of burning flame, we combine spatial and temporal features to remove spurious fire-like regions. Finally, some missing fire regions are located using region growing method. Experimental results indicate that the proposed approach can be applied to a variety of conditions and outperform some existing technique.