A video-based smoke detection algorithm using smoke color mixture model and multi-feature combination is proposed.The algori thm first employs mixture Gaussian background model to obtain the motion region of video,and then combines it with smoke color hybrid model to extract suspected smoke regions. Finally,the combination of smoke′s dynamic features (mean velocity and varianc e of moving velocity,motion direction,area growth rate) is used to identify smoke and exclude non-smoke from the suspected smoke regions.The experimental results conducted on the different types of smoke videos show that the average detection rate of th e proposed method for thick smoke with normal illumination is 96.57%,while that for thin smoke or smoke with weak illumination is 87.67%.The algorithm outperforms other state-of-the-art algorithms and has a wide range of applications.