Abstract: In order to improve the fire detection rate based on video monitoring in large-span space buildings,this paper proposed a fire recognition method based on LBP and GBP features with multi-scales.Series flame of fire images were preprocessed in RGB space at first,and the flame candidate areas were located by the stroboscopic feature.We established the Gaussian difference scale space for fire images,and then LBP-feature and GBP-feature with different scales were extracted from these candidate areas.Finally,these features were put into SVM classification to recognize whether it is a flame.Experimental results show that the combination of LBP and GBP is invariant to uneven illumination,and improves the accurate of recognition flame.

Key words: Multi-scales,Local binary pattern,Global binary pattern,Fire detection,SVM