A New Smoke Detection Method of Forest Fire Video with Color and Flutter

摘要

Big region area, long distance, and fast motion are the main characteristics of smoke video. Traditional smoke video detection has a high false alarm rate. In order to improve the detection accuracy, we propose a novel method to detect smoke video based on color information and flutter analysis. The proposed method can effectively exclude the influence of those objects resembling the smoke motion and reduce false detection rate resulting from illumination change and smoke-color objects. In detail, a background updating model of forest fire smoke video is first built using the Kalman filtering method. Combining updating model with the smoke HSV and RGB color space, candidate smoke video region is segmented. Then, three flutter features, including flutter direction of the video smoke, changing rate of the smoke area, and background ambiguity, are extracted by sliding time window analysis of the candidate smoke. Finally, the characteristic values are used to judge whether it is smoke. Experiment results show that our method can detect the video smoke more accurately as well as faster than the state-of-the-art methods.