Video Fire Recognition Research Based On Codebook

In order to improve the detection accuracy, a method using Codebook background model was proposed. The method which combines with, static and dynamic features of fire was innovatively applied with YUV color space in Codebook background model to detect fire region, and update the background regularly. First, extract frames from video, use the liner relation between R/G/B component as the color model to get the fire color candidate area.Second, convert color space from RGB to YUV, thus use background learning and background difference to obtain the candidate areas which are moving. At last,train BP (Back Propagation) neural network with the features vectors such as fire area change rate,fire area overlap rate,circularity and fire centroid displacement. Use the trained BP neural network to judge whether fire exists in videos or not.Choose fixed camera videos as experiment resource. When dealing with complicated video scenes, detection accuracy of the proposed algorithm can reach 96%. Experimental results show that compared with three state-of-art detection algorithms, the proposed one have higher accuracy in different fire scenarios and lower rate of misclassified fire frame in non-fire scenarios.