Flame Detection Method Based On Video

Fire is one of most the common serious disasters, It endangers the human life and property, and causes the environmental pollution, even ecological balance. At present the domestic and foreign all fire auto-alarm technology is mainly based on sensor. But in big space such as outdoor warehouse and large-scale indoor warehouse, the sensor' s above signal becomes weaker because of the huge space. Even if the high accuracy sensor is unable work because of all sorts of interference noise.The fire-detection algorithm based on video processing uses computer vision, pattern recognition and many kinds of technologies to analysis image information. It is not limited by distance, and automatically detects fire, and sounds the warning promptly to the system administration personnel. This paper implements the fire detection research using the ordinary camera and deeply study the flame characteristic and the detection algorithm.This paper proposes the thought of the moving target detection was applied to flame detection and regards the flame as a special moving object to extract the flame target by integrating the moving object detection and the flame' s visual characteristic. First, we use mixture Gaussian model to detect moving target including the flame, then analysis the moving target' s color to gain the region which has the flame' s color, many disturbances of the flame color static object are removed.Second, We analysis the flame characteristics and design the correspondent algorithm:1. Flame overall move detection.2. Flame area growth detection. 3. Flame region acute angle detection. 4. Flame high change detection. 5. Flame profile similarity detection. 6. The above characteristics are classified by BP neural networks to judges whether it is the flame or not.At present the majority flame detection methods need to judge whether it is conformed to the flame' s multi-characteristic or not, It is considered not to be the flame if a characteristic does not match the flame characteristics. Because of the flame' s multi-characteristic, it will lead to the omission. This paper uses BP neural networks to detect fire, if a flame characteristic not been examined, it will not lead to the omission. Simultaneously the system has the functions of the self-learning and adaptive capabilities. This method cuts the false alarm rate effectively.The experiment proved that the method of the moving target detection was applied to flame detection and a neural network to identify characteristics of the flame classification has higher identification and with good prospects for development.