Combining Gaussian Mixture Model And Hsv Model With Deep Convolution Neural Network For Detecting Smoke In Videos

Research on video analysis and processing of fire and smoke detection has gradually become a popular topic in computer vision. It is also a challenging task to detect smoke in videos due to the non-rigid characteristics and the large variance of smoke color, texture, shape, density and lighting, causing most of existing video-based smoke detection algorithms with high false detection rate. In this paper, we combine Gaussian Mixture Model (GMM) and HSV color model with the deep convolution model for detecting video-based smoke, which aim at filtering out no-smoke blocks to further reduce false detection rate and improve the detection accuracy. We evaluate our approach on many smoke video clips and demonstrate reduction of false detection rate and the improvement of precision.