提出了一种基于非参数特征提取的早期森林火灾烟雾检测方法。采用双背景模型进行运动区域检测，得到可疑烟雾区域。再利用早期火灾烟雾颜色特征和运动特性对提取的可疑运动区域进行颜色判别，对增长区域以及上升区域进行检测，排除部分干扰目标。根据提出的非参数特征提取法，计算并科学地选取可疑区域的特征量，将其输入训练好的支持向量机(SVM)进行烟雾判别。对多段视频进行测试表明，该方法能适用于复杂多变的野外环境，排除干扰，有效地检测出早期林火烟雾。

An early forest-fire smoke detection method based on non-parametric feature extraction is proposed. A dual background program is taken to detect the motion area to get the suspicious smoke area. The non-smoke moving targets are excluded from suspicious region according to the color features and dynamic characteristics of early smoke. In order to select the feature vectors scientifically, a non-parametric feature extraction is used to extract feature parameters in the candidate regions. These feature parameters are used as inputs of a SVM to distinguish between smoke and non-smoke. Experimental results on test videos indicate that the proposed method can achieve better accuracy and fewer false alarms compared with the state-of-the-art technologies and can be used in outdoor environment.