Multi-feature Detection Algorithm Of Video Smoke

In order to monitor whether fires occur or not in the open environment, a smokedetection method for early fire based on video processing technology is studied. The fires areusually accompanied by smoke. Compared with traditional smoke detectors, using fire videosmoke monitor, additional fire information (such as burning position, combustion growth rate)can be obtained, which is conducive to fire prevention.Firstly, three commonly used methods in moving target detection are studied, andbackground subtraction method based on improved running-mean method is determined. Themotion area in the current video frame in a video sequence of moving targets is acquired bybackground subtraction, and the motion area is processed by the regional holes filled andcommunication domain analysis method to obtain relatively accurate motion foregroundregion.Three features of smoke (color features, obscured features and texture features) arefound for the known suspected smoke area. The eigenvalues of the smoke color RGB threechannels are extracted through analyzing smoke color feature model of the combustibles. Theeigenvalues of obscured features model are extracted and analyzed using wavelet transform.Energy and contrast of texture eigenvalues are gotten by analyzing the image texture of themoving region. Through analyzing each type of smoke features, it is found that single featureamount of smoke detection reliability should be strengthened.Then, in order to further improve the accuracy of the smoke detection, the threefeatures(color features, obscured features and texture features) are extracted, which areregarded as non-linear discriminant based on support vector machine. Through the fusion ofmultiple determination features, the smoke detection accuracy is improved.Finally, to verify the effectiveness of the proposed algorithm, a number of differentscenarios videos with smoke, non-smoke and disrupting smoke are tested. The results showthat the video smoke detection algorithm in this thesis can identify smoke in video accurately, real-time and effectively, with anti-jamming capability.