Research Of The Flame Detection Algorithm Based On Video

With the rapid development of modern society, the density of population and buildings unceasingly increases, and the destruction of fire is rising year by year. Therefore, the technology of fire detection is receiving more and more attention of scholars both are domestic and overseas. Meanwhile, video monitoring system are widely available in transportation, finance, public security, etc, it also matting a broad market prospect for the application and promotion of new fire detection techniques, and the flame detector recognition is the most critical part of the whole system.The flame could be affected by many factors such as burning material, condition of illumination, air convection, interference of objects, so that presenting diversity and instability of visual features. In this thesis, the research based on the perspective of image analysis:firstly, based on movement characteristic and color characteristics of the flame, the video could be processed and the area of suspected fire could be extracted; then the Beacon information of edge textures which match to the edges of these regions.are calculated; On the basis of this, with the methods of scintillation statistics and others, robust features of the flame are extracted. Furthermore, with filling the flame area, which employed the seed growth method, a good basis of the post-identification is established through the seed growth method.For the video which is difficult to identify, the recognition rate under different kernel functions have been studied, which is combined with the theory of support vector machine, and the video materials of false alert on flame were collected and analyzed, which were based on the interference factors that producing false alert. Then, the available eigenvalue are extracted by using the methods of scintillation statistics, then the statistically recognition rate could be analyzed under the different the kernel functions.Lastly, this thesis presents the device of the experimental platform of flame detection system software, and tests the algorithm. In conclusion, this system has an excellent effect in the test.