Study On Fire Detection Method Based On Video Image

There are more and more fires around the world with the increasing number of high-rise buildings, and the loss caused by which is also growing. In order to reduce the loss and damage caused by fire as much as possible, determining and forecasting the fire accurately in the early stage to protect people's lives and property safety are very important and necessary.At present, the detectors of smoke, temperature or photographic are widely used in the general constructions. They detect fires by determining the signals of smoke density, temperature and photographic. These detectors are vulnerable to the limitation of space. And the sensitivity of outdoor detection in open space is low. With the rapid development of digital image processing technology, video-based fire detection system can overcome the shortcomings of environmental space constraints exist in early fire detectors, and has become the mainstream research direction in the field of fire.As the fire in video monitoring mainly has two obvious products:flame and smoke, the fire detection methods based on the characteristics of flame and smoke are analyzed deeply in this paper.(1) In the study of fire detection method based on flame features, the main research work and achievements are as follows:(a) On the basis of image processing technology, for the flame has features of changing slowly and flashing continuously, the experimental results of the flame motion region extracted from the background estimation method and the three difference method are analyzed. Result shows that three difference method can separate the flame region more accurately and adequately. So it's more suitable for flame region segmentation;(b) Depth analysis of flame in various color spaces are made in this paper. And it can be found that the Ohta color space is most suitable for dividing flame color. For different scenarios with different flame color range and the hole problem in motion area exacted from three difference method, Otsu algorithm is applied to select the flame color segmentation threshold automatically in Ohta color space. And the region growing algorithm is used to make up the hole problem. So the area with flame color can be segmented accurately. Finally, combined with the motion area exacted from three difference method, the motion area only with flame color is obtained. This method proposed can eliminate the stationary object with flame color and reduce the target region effectively.(c)The comprehensive features of flame (such as the Wavelet energy, stroboscopic, contour and the number of connected regions features) are combined to further determine whether there is fire or not. And according to the ability of each feature to identify flame, a weight is set to each feature. And whether there is a flame or not in suspicious areas can be determined. Contrary to the fire detection method based on flame features proposed in this paper, some simulation experiments have been done. Experimental results show that the fire detection method can be applied to both indoor and outdoor and complex background environment with a good anti-interference ability and high recognition accuracy.(2) In the study of fire detection method based on smoke features, the main research work and achievements are as follows:(a) According to the problem caused by slow diffusion of smoke that background update in background estimation method is not very accurate, an improved background estimation method is proposed in this paper. That is, the original background is introduced to the formula of background update. At the same time, the background estimation threshold is also updated with the updating of background. An accurate motion area can be extracted with this method. This can lay a foundation for the further smoke detection.(b) As the shortcomings of a fire detection method just with single smoke feature, such as the high rate of false and lose alarms, the smoke integrated features (including color, contour, fuzzy background and diffusion features) are analyzed to detect whether there is smoke or not in a motion area. Contrary to the fire detection method based on smoke features proposed in this paper, some simulation experiments have been done. Experimental results show that this algorithm proposed is not restricted by environment space. It can be applied to both indoor and outdoor environment and can exclude disruptors successfully with high recognition accuracy.