基于视觉显著性和小波分析的烟雾检测方法  

Vision Saliency and Wavelet Analysis Based Smoke Detection

作　　者：秦文政[1] 马莉[1] 

机构地区：[1]杭州电子科技大学生命信息与仪器工程学院,浙江杭州310018

出　　处：《杭州电子科技大学学报（自然科学版）》2011年第4期114-117,共4页Journal of Hangzhou Dianzi University：Natural Sciences

基　　金：国家自然科学基金资助项目(60775016);浙江省科技计划资助项目(C13062)

摘　　要：基于视频图像的烟雾检测对火灾预警系统性能具有重要作用。该文提出一种基于时空域信息处理的烟雾检测算法:基于视觉注意模型的时空域ROI区域提取,以及基于小波时域分析的烟雾动态特征识别。该文用基于颜色对比度、亮度对比度和运动3个初级视觉特征的显著性图融合捕捉场景中的ROI区域。针对在时域中烟雾区域小波高频能波动小以及归一化颜色信息基本不变的特点,利用小波分析的ROI区域小波高频能下降率和颜色不变性特征通过贝叶斯分类器实现烟雾的识别。试验结果表明,该文提出的方法能有效去除类烟运动物体的干扰,检测准确性高。Video-based smoke detection is important to fire alarming systems. A method of smoke detection based on space-temporal information processing is proposed in the paper： region-of-interest is extracted using the model of vision saliency in the space-temporal field, and recognition of smoke dynamic features is implcmented by vavdet analysis ：in time domain. Three visual features in low level are fused for capturing the ROI.With eonSideiraation of tion for smoke＇ textures high frequency and little change for normalized color informa- implemented using a Bayesian classifier through both decreasing ratio of wavelet energy in high fi＇equency and color invariant feature by wavelet analysis. Experimental results show that the proposed method has strong robustness to interferences （such as moving objects with smoke-like color） w!th h!gh accuracy

关 键 词：视觉显著性 感兴趣区域提取 小波分析 烟雾检测 

分 类 号：TN401[电子电信—微电子学与固体电子学]