基于运动和亮度显著性的森林烟雾分割方法

A Forest Smoke Segmentation Method based on Saliency of Motion and Brightness

本文提出了一种基于运动和亮度显著性检测的烟雾区域分割方法,目的是解决传统的运动检测方法对于树叶抖动、摄像机抖动等不显著的运动区域比较敏感的问题。采用低秩结构化稀疏分解方法提取前景区域,然后计算烟雾的显著性,以便进一步分离。我们提出一种基于自适应参数的群稀疏鲁棒标准正交子空间学习(ROSL)的显著性测量方法。实验表明,该方法能够很好地处理大范围的烟雾视频,并能获得较好的烟雾检测结果。

This paper proposes a smoke region segmentation method based on motion and brightness saliency detection,aiming at problems that traditional motion detection methods are sensitive to insignificant motion regions such as leaf shake and camera shake.The low-rank structured sparse decomposition method is used to extract foreground area,and then smoke saliency is calculated for further separation.We propose a saliency measurement method for group-sparse Robust Orthogonal Subspace Learning(ROSL)by virtue of adaptive parameters.Experiments show that the proposed method can work well on a wide range of smoke videos,and obtain better smoke detection results.