Based On The Video Sequence Of Multi-feature Flame Detection Algorithm

The fire which is one of the main production disasters has been threats to human life. In the modern society, with the development of science and technology, the cause of the fire is gradually increased. How to present fire automatically and efficiently becomes a pressing demand. The traditional automatic fire prevention measures which is based on the detection of sensing instruments, sensing instruments used the flame of physical properties, such as smoke, temperature, light intensity and so on, through the perception of the physical characteristics of flame to get flame signals and call the police, because it adapts widely in the small space, becoming an important measure of the fire detection. At the same time the sensor also has inevitable disadvantange, such as when the fire area is far from sensing instruments, the sensor signals would be very weak and the cost of high precision sensors are very expensive, etc. So people in desperate need a fire detection technology with a wide detection area, accurate detection orientation and low cost.Video based fire detection is a set of computer vision, artificial intelligence, pattern recognition, signal processing technology and so on to analysis and process the flame video, not affected by distance, and higher precision. Because it uses small equipments, covers a wide range and cost relatively low, become the fire monitoring field new concerns and the development direction.In this paper, we extract the characteristics of the image of flame and analysis it in the spatial domain by our algorithm, constructing the recognition model on fire and realizing the flame characteristics through matching and detecting. First proposed an prospects detection method combined with Gaussian prospects detection method and flame color model can obtained video motion in the foreground area, after the movement of the foreground region obtained by the analysis of the characteristics of the filter, concludes the recognition model, compare with the standard model by experiment, and ultimately determine the identify results. Recognition model, mainly based on some basic character, including color features, pointed flame area feature, volatility features. The article is classified the characteristics as judge and validation features by the experiment, through the judge features to determine the suspected area of flame, used validation features to verify the region of flame.Volatility is the key attribute in the current flame detection algorithm of the extraction of flame characteristics, but most volatility only pay attention to the local information that is the volatility of the pixels, not considered the whole of the flame fluctuations. This paper proposes a flame detection method based on the algorithm of flame outer flame volatility it will extracts the features of the volatility of the region of outer flame and calculates, they provide better features for flame recognition. At the same time, build the characteristics of the flame established regional identification model, which reduce the error rate of fire detection and miss rate.Finally, experiments show that the flame detection methods provide a higher recognition rate and a lower error rate, it is better to recognize the flame.