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。

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

Abstract: 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.

Key words: Video Smoke Detection(VSD), image enhancement, saliency detection, Gaussian Mixture Model(GMM), smoke segmentation

中图分类号: 

• TP301.6