Abstract: This paper presents a smoke detection algorithm characterized by combination of serial and parallel processing model. This algorithm analyzes multiple features of video sequence by extracting the motion foreground of Gaussian mixture background modeling, and circling interested regions. When analyzing color feature, the information of each channel is normalized in RGB space, and the threshold value is judged according to the color characteristics of the smoke. When analyzing the shape feature, the statistical method is adopted to monitor the irregular degree of the video image with the measurement standard of irregular degree of break variable. Taken change rate of wavelet coefficient as check standard, the wavelet transform method is adopted to detect the high frequency information in the image with the combination of the characteristics of smoke diffusion. Based on the weighted analysis of multiple features, a comprehensive criterion is established for the detection and alarm of smoke detection in video image.

Key words:  smoke detection, serial and parallel processing model, Gaussian mixture model, color analysis, shape analysis, wavelet analysis, dynamic threshold