A Fire Flame Recognition Algorithm Based On Video Features

A fire flame recognition algorithm based on video features is presented in this paper. First, the RGB images are grayscaled by a standard averaging method, following an appropriate threshold is selected in binaryzation by the two-apex method, and then the suspicious regions are connected using a morphological close operation. With the regions labeling in consequence, characteristic parameters between target and interferences are calculated to identify a fire. Secondly, the static and dynamic features of fire flame are showed in this paper, along with Area, Mean Gray, Circularity and centroid of the fire and interference regions. It is found in the experiment data that the differences among the autocorrelation function of the area sequence, the variance of the mean gray sequence, the mean value and autocorrelative function of the circularity sequence, and the distance the centroid moves of the fire region and that of interferences is sensible. A proposed use of these five dynamic video features as parameters to identify a fire is described based on the above. Finally, the basic theory of statistical pattern recognition is described, along with Bayes classification. And then Bayes classifier is designed to identify the fire flames by the theory of smallest error Bayes decision.It shows in the experiment result that the dynamic video features and the recognition algorithm described in this paper can identify correctly fire flame, and proved to be able to exclude the interferences from single lightspots, blaze under control, flashlight and portable lighters.