Fire Detection Algorithms Using Multimodal ... - Bilkent University

CHAPTER 6. WILDFIRE DETECTION 82In this chapter, a computer vision based method for wildfire detection is presented.Currently, average fire detection time is five minutes in manned lookouttowers in Turkey. Guards have to work 24 hours in remote locations under difficultcircumstances. They may get tired or leave the lookout tower for various reasons.Therefore, computer vision based video analysis systems capable of producingautomatic fire alarms are necessary to reduce the average forest fire detectiontime.There are several approaches on automatic detection of forest fires in theliterature. Some of the approaches are directed towards detection of the flamesusing infra-red and/or visible-range cameras and some others aim at detectingthe smoke due to wildfire [22] [46], [4] [36]. There are also recent papers onsensor based detection of forest fires [39], [70]. Infrared cameras and sensor basedsystems have the ability to capture the rise in temperature however they are muchmore expensive compared to regular pan-tilt-zoom (PTZ) cameras.It is almost impossible to view flames of a wildfire from a camera mountedon a forest watch tower unless the fire is very near to the tower. However, smokerising up in the forest due to a fire is usually visible from long distances. Asnapshot of a typical wildfire smoke captured by a look-out tower camera from adistance of 5 Km is shown in Fig. 6.1.Guillemant and Vicente based their method on the observation that the movementsof various patterns like smoke plumes produce correlated temporal segmentsof gray-level pixels. They utilized fractal indexing using a space-fillingZ-curve concept along with instantaneous and cumulative velocity histograms forpossible smoke regions. They made smoke decisions about the existence of smokeaccording to the standard deviation, minimum average energy, and shape andsmoothness of these histograms [36].Smoke at far distances (> 100m to the camera) exhibits different spatiotemporalcharacteristics than nearby smoke and fire [84], [23], [87]. This demandsspecific methods explicitly developed for smoke detection at far distances ratherthan using nearby smoke detection methods described in [86]. This approach is inaccordance with the ‘weak’ Artificial Intelligence (AI) framework [61] introduced