Fire Detection Algorithms Using Multimodal ... - Bilkent University

CHAPTER 3. FLAME DETECTION IN INFRA-RED (IR) VIDEO 36for example, they may produce false alarms to reddish leaves flickering in thewind and reflections of periodic warning lights. IR cameras can be used to realizerobust systems. However IR cameras and systems are more expensive than regularcameras.A bright-looking object in IR video exhibiting rapid time-varying contours isan important sign of presence of flames in the scene. This time-varying behavioris not only directly observable in the contours of a fire region but also observableas variations of color channel values of the pixels in regular video. On the otherhand, entire fire region appears as a flat white region in IR cameras operating inwhite-hot mode.As pointed out in Chapter 2, turbulent flames flicker with a frequency ofaround 10 Hz [14] and [1]. Various other flame flicker values were reported fordifferent fuel types in [5] and [42], such as 11.7 Hz and 12.5 Hz. The flicker processis modeled using Markov models as in regular video. The use of infra-red (IR)cameras instead of a regular camera provides further robustness to imaging basedfire detection systems especially for fires with little radiance in visible spectrum,e.g. alcohol and hydrogen fires which are common in tunnel collisions. Unfortunately,the algorithms developed for regular video cannot be used in IR videodue to the lack of color information and there is almost no spatial variation orvery little texture information in fire regions in IR video as in most hot objects.Therefore, new image analysis techniques have to be developed to automaticallydetect fire in IR videos.In IR video, boundaries of moving bright regions are estimated in each IRimage frame. It is easier to estimate hot object boundaries in IR video to contourestimation in color video. A one-dimensional curve (1-D) representing the distanceto the boundary from the center of mass of the region is extracted for eachmoving hot region. The wavelet transform of this 1-D curve is computed and thehigh frequency nature of the contour of the fire region is determined using theenergy of the wavelet signal. This spatial domain clue replacing the spatial colorvariance information in regular video is combined with temporal clues to reach afinal decision.