森林背景下基于自适应区域生长法的烟雾检测

Smoke detection based on adaptive region growing method in forest background

森林背景下,有效的烟雾检测在避免大规模森林火灾方面具有极其重要的意义。当前的研究对烟雾移动得很慢或没有清晰背景的情况下往往表现较差的性能,提出一种针对烟雾检测的自适应区域生长法。采用改进的卡尔曼滤波检测出运动区域,假设烟雾的亮度与视频照度之间存在线性关系,采用支持向量机(support vector machine,SVM)线性回归方法得到烟雾亮度的近似范围,并定义亮度约束,基于检测得到的运动区域,同时考虑亮度约束和纹理约束,蔓延出烟雾区域的主要部分,提取基于区域的特征来做SVM分类。对比实验结果表明,该方法优于传统的方法,并具有更强的鲁棒性。

An effective smoke detection is very important to avoid large-scale forest fire in forest background. The current study has poor performance in the following situations where the smoke is moving very slow or there is no clear background.In order to solve these problems,a specific smoke detection with adaptive region growing method is proposed. Firstly,we use the improved Kelman filtering to detect the motion region. Secondly,the contact between smoke brightness and intensity of illumination is assumed as a linear relation,and we use SVM as a linear regression to obtain the approximate range of smoke brightness,and define this as a luminance constraint. Then we can spread out the main part of smoke region by considering the combination of luminance constraint and texture constraint based on detected motion region. Finally,some features based region has been extracted for SVM classification. Experimental results show our method outperforms the conventional methods,and have more robustness.