Abstract: Towards the requirement of real-time smoke detection in DSP surveillance system, a novel fast video smoke detection algorithm is proposed.Compared with current video smoke detection algorithms, new algorithm does not demand storing historic frame images of video sequence, and does not require learning background image of monitoring scene.According to extracted fluttering characteristics of smoke for independent image block, Bayesian decision method is used to determine whether smoke is existent.Derived from above design, new algorithm only requires very slow computational complexity.Nevertheless, benefited from a group of effective video smoke characteristics presented in this paper, new algorithm also earns high detection performance.Accordingly, new proposed video smoke detection algorithm is very valuable for real-time smoke detector embedded in a standard video surveillance system.