B017 Fire image texture

New fire criterion based on texture from video images

Brief:

In order to increase the exactness of early-fire protection based on video sequences, a new fire criterion based on texture of video images is proposed in this paper. The four main statistical textural characteristics of consecutive frames from four directions are calculated and the maximum changing rate of contrast of the consecutive frames in the each direction is presented, which can be used to discriminate fire from fire-like objects no matter whether the objects are motional or stationary. In addition, a new segmentation method of video images is developed based on entropy, Otsu, region growing, et al, which can successfully segment fire-like objects from the original image sequences in different environment, different background and different form. Therefore, the textural matrices are computed only on these objects and the calculating time is cut down greatly. This criterion is verified through many experiments of fire video images and fire-like video images. Experimental results indicate that the contrast is a valuable description amid the commonly used textural features and it may be regarded as fire if its average contrast variety of four directions is greater than 80%. The method proposed can be used as a supplementary criterion in automatic fire detection using video images.

Further information:

Li J, Fong NK, Chow WK, Wong LT, Xu DG, Lu PY, 2004. New fire criterion based on texture from video images, 3^rd International Symposium on Instrumentation Science and Technology, August 18-22, Xi’an, China, pp. 2/0171-2/0176.