Abstract：In order to solve the problem of an early fire detection in large spaces and to reduce the impact of environmental changes on the forecast,this paper explores an approach for a video fire smoked detection based on analyzing the visual characteristics of smoke images. It includes the three parts. Firstly,the difference image is attained by using the background subtraction method. Then,the suspicious regions are obtained by the binary image method and mathematical morphology processing. Finally,the color features,motion feature and shape feature extracted from the suspicious regions are detected based on Adaboost and RBF neural network to judge whether there is smoke in the scene or not. The experiment results show that this method can detect smoke effectively and has good anti-interference capability,as well as can improve the accuracy of smoke detection. It has good application prospects.

Keyword：background subtraction;features extraction;RGB normalization;AdaBoost;RBF neural network;