Fire Detection Algorithms Using Multimodal ... - Bilkent University

CHAPTER 4. SHORT RANGE SMOKE DETECTION IN VIDEO 554.1 Detection AlgorithmThe flames of a fire may not always fall into the visible range of the cameramonitoring a scene covering large areas like plane hangars or open spaces. Firedetection systems should tackle with such situations by successful detection ofsmoke without flame. In this chapter, temporal and spatial wavelet analysis aswell as an analysis of contours of possible smoke regions are carried out for smokedetection.Smoke gradually smoothen the edges in an image. This characteristic propertyof smoke is a good indicator of its presence [52], [10]. Edges in an imagecorrespond to local extrema in wavelet domain. Degradation of sharpness in theedges result in a decrease in the values of these extrema. However, these extremavalues corresponding to edges do not totally boil down to zero when there issmoke in the scene. In fact, they simply loose some of their energy but they stillstay in their original locations, occluded partially by the semi-transparent smoke.Independent of the fuel type, smoke naturally decrease the chrominance channelsU and V values of pixels. Apart from this, as discussed in the previouschapters, both flame and smoke are turbulent phenomena. Smoke regions havetime-varying boundaries similar to flame regions. Smoke partially covers anduncovers background objects especially at the early stages of fire. Therefore, aMarkov model based modeling of turbulent smoke behavior is appropriate as inflame detection. However, smoke boundaries move with a lower frequency at theearly stages of fire.In addition to color and turbulent behavior analysis, boundaries of smokeregions are also estimated in each video image frame. A one-dimensional curve(1-D) representing the distance to the boundary from the center of mass of theregion is extracted for each smoke region. The wavelet transform of this 1-Dcurve is computed and the high frequency nature of the contour of the smokeregion is determined using the energy of the wavelet signal. This spatial domainclue is also combined with temporal clues to reach a final decision.