Real Scene Flame Detection Based On Video

With the development of video monitoring and digital image processing technology, and the limitation of traditional application of fire detector, the indoor and outdoor detection algorithm based on video has been put forward by an increasing number of people. However, the current detecting algorithm is only valid in a certain space such as experimental environment because of the complexity of fire conditions, the false rate in the natural environment is still in a high level. The detecting algorithm present in this paper can adapt to the changing natural environment and accurately alarm for the fire flame in an early period, thus greatly improve the detection rate of fire.At present, most flame detecting algorithms are all analyses based on the color and movement of fire in the videos. And all the detecting algorithms are focus on the daytime fire detecting or smog detecting. But when the fire disasters happen at night, the detecting algorithms would fail to get the real color of the fire for the far distance and the differences of burning objects and temperature. Therefore, the color of fire can not act as the feature of fire identification. What we need is a effective way to detect the fire in long distance as well as the fire at night. The detecting algorithm put forward in this paper effectively solved this problem since it can detect the fire only through the gray-level information and the movement of fire.The algorithm contains six parts: Flame background modeling, background update mask creating, detecting of slow moving objects, highlight area detecting, suspected flame area extraction, regional marker tracking and identification.According to the stabilization of fire in early period, the author adopts the advanced VIBE algorithm to establish the background modeling, and with the application of background update mask, the fire would adapt to the changing natural environment without being updated on the background. Therefore, the accuracy of identification will be improved by the exact extracting of the flaming area.Through a large number of experiments,it is found that there are no obvious sharp features of the fire in early period, and to identify the fire by the circularity is also difficult. Thus, based on the analysis of the dynamic feature of fire, the author put forward the method of identifying the fire by the change intervals of scale, gray level and central position etc.