Research On Fire Detection Based On Smoke Image Recognition

The sudden and strong fire,the threat of people's lives and property safety,how to accurately detect the beginning of the fire has been the focus of people's research.With the development of smart city,video surveillance technology has been widely used in various fields of life,the use of video surveillance technology instead of the traditional fire detector has become a new trend of fire detection.In the development of fire,the fire will be a period of time after the smoldering.So,compared to the flame,smoke appears earlier,is an important symbol of early fire.This paper mainly studies the fire detection method based on smoke image recognition,firstly,Gauss mixture model is used to extract the smoke motion region in video sequences.then the color characteristics,morphological features,area growth characteristics,motion direction features and texture features of fire smoke are analyzed;in the video image,the water mist is very similar with fire smoke,few people to study on the interference of water mist in fire smoke detection,In this paper,the characteristics of fire smoke and water mist in the same scene are compared,and find the different characteristics of fire smoke and water mist in video images.Finally,using support vector machine training and recognition of video.The experimental results show that when the video set is the fire smoke and water fog,use the direction of motion features and texture features can distinguish the fire smoke and water mist,which proved that fire detection method in this paper can't be interference by water mist;when the video set is the fire smoke and none fire smoke,the identification rate of fire smoke is calculated,and the false alarm rate of non fire smoke is calculated,And compared with other methods,it is demonstrated that the method can accurately identify the fire smoke at the same time,also can greatly reduce the false alarm rate caused by suspected smoke target,make the fire detection more accurate.