Stereo-Vision-Based Fire Detection and Suppression Robot for Buildings

Background

In recent times, research on the detection of fire flames using surveillance cameras with machine vision has gained momentum. The image processing approach involves the extraction of the fire flame pixels from a background by using frame difference technologies. Healey et al. (1993) presented a fire detection algorithm using a color video input with a pre-partitioning scheme under some restricted conditions, without rejecting the similar fire-like alias. Phillips et al. (2002) and Celik et al. (2007) conducted studies on computer vision by using spectral analysis and the flickering property of fire flame pixels to recognize the existence of fires at a scene. Hue and saturation are adopted as feature vectors to extract the fire pixels from the visual images (Chen, 2003). Fire flame features based on the HSI(hue, saturation, intensity) color model are extracted, and regions with fire-like colors are roughly separated from the image by the color separation method (Horng, 2005). Then, the image difference method based on chromatics is used to remove spurious fire-like regions such as objects with similar fire colors or areas reflected from fire flames. A fuzzy-based dominant flame color lookup table is created, and fire regions are automatically selected (Wang, 2006). However, either the solution does not consider the temporal variation of flames or the approach is too complicated to process in real time.