Study Of Fire And Smoke Detection Method In Video

The use of fire is a great creation of human society, and it plays an immeasurable role for social development and progress. The fire benefits humankind if people can grasp the laws in various fire places, improve the vigilance of using fire and take a reliable means of prevention. On the contrary, once the fire flame is out of control in time or space, it will lead to a fire which often endangers human life, destroys natural environment, damages material wealth and causes huge losses to human society. The fire is the most frequent and the highly destructive one in various types of disasters.Struggling tenaciously with the fire for many years, people got to know some laws of fire formation and development, and they also accumulated a lot of experience, methods and measures of prevention of fire. With social development and technological progress, people use a variety of fire detection methods and techniques to detect fire. Traditional fire detection methods are based on temperature detectors and smoke detectors. When the detection value exceeds the set threshold, the detector will alarm. This fire detection method is the most widely used and the most mature technology. However, they are not suitable to detect fire for large space buildings and open areas. Currently, there exist some detection technologies based on infrared and ultraviolet. However, these kinds of devices are relatively expensive. Video surveillance systems are popularized in the city and fire key protection units now. With the development of artificial intelligence and pattern recognition technique, fire detection method based on video is attached more and more importance. Comparing to the optical detection equipment, the price of CCD and CMOS is lower. This method can also use the existing video surveillance equipment and even some hardware equipment is not necessarily purchased, which can reduce the cost of fire detection system. It is also suitable to detect fire for large space buildings and open areas.According to the motion characteristics of early smoke, a swaying object detection algorithm was presented in this paper. According to the swaying and diffusion feature of smoke, a method of early smoke detection in video using swaying and diffusion feature was presented in this paper. According to the motion, pixel color and area variation feature of fire flame, a flame detection synthesis algorithm was presented in this paper. Research on unique edge shape of smoke and flame contributes to fire detection. Inspired by this idea, a new algorithm based on growing cell structures to realize curve reconstruction using line segment was presented. Flame and smoke edge curves will be reconstructed in the next step. The main research results are as follows:1. A swaying object detection algorithm was presented. Some moving objects such as early smoke and flame have unique vision features. Their bottom region is less mobile than their top region. This motion mode is called swaying feature in this paper. For this swaying feature, a swaying object detection algorithm was presented in this paper. Firstly, fuzzy integral is adopted to extract moving objects from video frames. Secondly, a swaying identification algorithm based on centroid calculation is used to distinguish the swaying object from other moving objects.2. Early smoke in video has swaying feature. According to the physical motion characteristics of gas, smoke is diffused as it is influenced by buoyancy of the air and eddy. Generally speaking, early smoke in video spreads towards upward, left upper side or right upper side. Smoke bottom region is less mobile than top region and smoke upper position is less stable than the lower position. This motion feature of smoke is called swaying feature in this paper.3. A method of early smoke detection in video using swaying and diffusion feature was presented. When an early smoke event occurs, smoke in video has unique vision features, namely swaying and diffusion. Smoke bottom position has relatively a small change, correspondingly to a relatively small change for centroid position of smoke bottom. Smoke top position has relatively a big change, correspondingly to a relatively big change for centroid position of smoke top. Therefore, a swaying feature is shown. Smoke has a continuous diffusion feature. In the bottom of smoke, smoke concentration is higher and the color basically shows smoke color. In the top of smoke, smoke concentration is lower and the color appears to be blended with smoke color and background color. In other words, the texture is comparatively rough in smoke bottom region and the texture is comparatively silky in the top region. Based on the two features of smoke, a method of early smoke detection in video using swaying and diffusion feature was presented in this paper. Firstly, choquet fuzzy integral is adopted using each component in YCbCr color model to extract moving regions from video frames, and then, a swaying identification algorithm based on centroid calculation is used to distinguish candidate smoke region from other moving regions. Secondly, smoke diffusion makes different textures between bottom region and top region of smoke. Gray Level Co-occurrence Matrixs are used to differentiate smoke from other candidate smoke regions.4. A flame detection synthesis algorithm was presented. Firstly, choquet fuzzy integral is adopted to integrate flame color features and texture feature for extracting moving regions from video frames. In YCbCr model, red and brightness components are independent with each other. The algorithm integrates the two components and local binary pattern texture for extracting moving object, which effectively avoids the interference of brightness variation in the video frames. Secondly, mean filtering is used to smooth RGB value of video frame pixels and detected moving regions are filtered by a flame color filtering algorithm to extract candidate flame regions. Finally, a flame area variation identification algorithm is used to distinguish true flames from candidate flame regions.5The edges of smoke and flame have unique shape. Extracting and analyzing edge curve feature of flame and smoke contribute to fire detection. An algorithm based on growing cell structures to realize curve reconstruction using line segment was presented. Given a set of unorganized data points and an initial polygonal line,the vertex position of polygonal line can be optimized by using the algorithm to make the vertexes of polygonal line gradually approach the given unorganized data points. In order to make the vertexes of polygonal line distribution coincide the space distribution of unorganized data points, the very active vertexes are split and the least active ones are deleted continually. This algorithm lays a foundation for edge curve reconstruction of flame and smoke for the next step.