This article provides a novel method to recognize fires in areas monitored by a color video camera. Formulated with a graphical programming language, the developed software utilizes the HSI attributes to extract, statically and dynamically, the pixels of flames with high intensity (I value) as well as those of peri-flame regions within a specific range of saturation (S value). The other pixels that are filtered out are regarded as the backgrounds. Some sample clips of fires and pseudo fires are processed with the software in which the high intensity and the specific saturation regions can be effectively segmented from the images. For each fire clip, after analyzing the areas of foreground pixels for both I value and S value along the time axis, there exists a high correlation between both sequences. The pseudo fire clips, on the other hand, do not demonstrate high correlations after the same processing. Therefore, a fire incident can be identified according to the correlation of both extracted pixel areas. This promising achievement has laid down a firm basis for the development of a novel fire detecting alarm system in the near future.

No abstract