Early smoke detection of forest fire video using CS Adaboost algorithm

Smoke is an important indicator of forest fire video. However, many distractors such as heavy fog and smoke-like moving objects greatly degrade the recognition accuracy. This paper presents a novel smoke detection method of early forest fire video using CS Adaboost algorithm. First, motion regions are extracted from two adjacent frames by a proper background model which can avoid false positives of some static distractors, such as blue sky and gray leaves. Then, a CS Adaboost algorithm is used to recognize smoke regions using centroid movement by means of smoke flutter, image energy on the basis of the Wavelet Transform coefficients and color information between a reference smoke color and the input frame. Finally, the experimental results show that the proposed method can not only detect smoke image of early low thickness, but also more effectively distinguish dense fog from smoke.