Video Flame Detection In Complex Scene

With the development of society, people are increasingly adapted to the use of fire in both life and industry. However, the frequency of fire disaster is also growing. In order to make life more safe, people began working to improve the security and protection equipment. Fire detection technology is also receiving much more attention, and it has become an important research topic at home and abroad. In recent years, video surveillance technology is widely used under transportation, residential and warehouse environments. Making full use of video surveillance will help to detect and analyze fire information in video, and it will raise the alarm timely and provide effective protection of personal safety and property security.Video flame detection technology is an important basis for the identification of fire, and it can determine whether there is a fire through flame features. In fact, because of the combustion of different materials influenced by light, air and other factors, the flame will show the diversity of different visualization features, such as motion, color characteristics, flicker, etc. This paper studies these features of flame, and then designs video-based fire detection algorithms.In motion detection, this paper analyzes present motion detection algorithms in video and compares their advantages and disadvantages. Moreover, aiming at the changes of flame block area in video scene, this paper detects motion area by using blocks in spatial consecutive video frames. This motion detection algorithm can improve correlation between adjacent pixels and reduce noise interference. Through this algorithm can quickly navigate to a suspected flame area, which greatly reduce the detection time and improve the detection efficiency. Then this paper designs two video flame detection methods:One is designed for video flame detection based on color block area moments. The method does some statistical analysis on a large number of color moments in the video flame block area, gets color variation of moments of flame block area, and then builds their color moments of color model for processing. By checking the video frames with color moments model, it can get suspected flame zone. Then it utilizes a different frequency between flame flickering pixels and non-flame pixels, and uses wavelet transformation to detect the frequency characteristics of the block area. Finally, locally weighted operator is used on flame detection area, which can eliminate single non-flame pixel’s interference.The second is designed for video-based flame detection method. This method detects a single pixel in HSV color space. Using hue, saturation and brightness to establish pixel color model, this method reduces brightness interference. Then this method uses temporal wavelet transform to analysis flicker characteristics of pixels and uses spatial wavelet transform to analysis flame block area of energy, which can determine the suspected area. Finally, this method excludes non-scattered filtered pixels to improve the accuracy rate of the video flame detection algorithm.Experiments show that the two methods can effectively detect the flame area in video and present the alarm area in image in time.