基于显著性检测和高斯混合模型的早期视频烟雾分割算法  

Early Video Smoke Segmentation Algorithm Based on Saliency Detection and Gaussian Mixture Model

作　　者：贾阳[1] 林高华 王进军[1] 方俊[1] 张永明[1] 

机构地区：[1]中国科学技术大学火灾科学国家重点实验室,合肥230026

出　　处：《计算机工程》2016年第2期206-209,217,共5页Computer Engineering

基　　金：国家自然科学基金资助项目"民航机舱特殊环境火灾图像特征规律及视频探测方法研究"(U1233102);国家"973"计划基金资助项目"城市高层建筑重大火灾防控关键基础问题研究"(2012CB719702)

摘　　要：针对视频火灾探测中早期火灾烟雾提取问题,提出一种基于显著性检测和高斯混合模型的烟雾疑似区域分割算法。根据人眼视觉注意机制,将阴燃烟雾看作视频中湍流和灰色显著的区域,采用显著性方法分割疑似烟雾区域。使用非线性增强方法增强视频的亮度图像和光流图谱,用增强后的图像计算显著性谱。由计算出的运动前景构造运动能量函数,对显著性谱进行估计,得到疑似烟雾区域。实验结果表明,与传统烟雾区域检测算法相比,该算法具有更好的分割精度,并且计算速度也有较大提高,适用于实时视频烟雾探测。An accurate candidate smoke region segmentation method using saliency-based method and Gaussian Mixture Model（ GMM） for Video Smoke Detection（ VSD） is proposed. According to Visual Attention Mechanism（ VAM）,smoldering smoke can be seen as a salient region in a video with its turbulence behavior and grayish color. A saliency-based method is used to do candidate smoke region segmentation. Saliency map is calculated using the nonlinearly enhanced luminance image and motion energy map. Suspected smoke regions are segmented from the saliency map using a motion estimation function calculated with GMM. The results are compared with those of three other methods used in the literature,revealing the proposed method has both a better segmentation result and better precision,and much faster operation speed than the existing methods. The proposed method is appropriate for real time VSD applications.

关 键 词：视频烟雾探测 图像增强 显著性检测 高斯混合模型 烟雾分割 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]