Abstract: Fire detection based on video image is an important research way to fire prevention. In order to improve fire detection rate, this paper formulates a novel color model to extract the candidate fire area with the color information of RGB and HSI color space at first. Then this paper proposes a novel method to extract the flicker frequency feature based on accumulative difference of three-channels of RGB color space, analyzes the flicker frequency feature values by logistic regression (LR) so as to get the optimized weight and slope, and establishes the probability model of flicker frequency feature value based on the optimized weight and slope. Finally, this paper uses the probability model in fire detection. The experimental results show that the proposed method makes the fire detection high accuracy and wide application, especially superiority for detecting the small flame.

Key words: fire detection, fire flicker frequency, color model, logistic regression, accumulative difference

摘要： 基于视频图像的火焰检测是火灾预防的一个重要研究方向。为了提高火焰的检测率，利用RGB和HSI色彩空间中的颜色信息，建立了一种新的火焰色彩模型，应用该模型提取疑似火焰区域。提出了一种基于累积差分RGB三通道的火焰闪频特征抽取方法，并用逻辑回归（logistic regression，LR）对火焰的闪频特征进行分析，得到了优化的权重和偏斜率，建立了火焰闪频特征值的概率模型。最后将概率模型应用于火焰检测。实验结果表明，该算法对火焰区域检测效果好，适用范围广，且能检测出较小的火焰区域。

关键词: 火焰检测, 火焰闪频, 色彩模型, 逻辑回归, 累积差分