Fire Flame Detection Algorithm Based On Monitoring Platform

Fire plays an important role in our life, but the event of fire, the consequences would be disastrous, in order to reduce losses caused by fire, the fire should found in the early stages, the fire is nipped in the bud, but the existing fire alarm devices in actual use certain limitations, easily affected by the interference of the external environment, it cannot be achieved in early fire alarm. In order to overcome the deficiencies of the fire detectors, this paper proposes an indoor fire detection algorithm based on video surveillance platform, which can effectively overcome the shortcomings of existing fire detection equipment,can send the alarm signal in the early fires, minimize losses caused by the fire.This study based on the existing video surveillance platform, first get the SDK development kit supplied by the video surveillance manufacturers, get the video stream through the SDK development kit, then convert the video code stream into a format convenient for video processing, while using the way of shared memory to achieve the video data between different processes.The next main research the flame foreground extraction algorithm, including frame difference method, Gaussian background modeling method and the optical flow method, in-depth study on the basis of these algorithms, combined with the flame movement characteristics and color information, we propose a new type of flame prospects extraction algorithm, by experimental we verify this prospect by extracting algorithm can obtain the full flame contour, facilitate the feature extraction in subsequently fire detection. Based on access the complete outline of the flame, we deeply research the main features of the flame recognition under different light intensities, including static characteristics and dynamic characteristics, and flame feature extraction algorithm, ultimately we decide to use the color characteristics, frequency characteristics, relevant characteristics, roundness characteristic feature of upward flame as the main feature recognition.Finally, a large number of experiments show that this system can accurately identify a variety of different fire burst in indoor environments, and can effectively avoid the error alarm caused by the light moves, silhouette shaking interference, the experimental results show that the recognition accuracy of this algorithm in above 90%, the recognition time around 5 to 20s.