摘要
烟雾是林火发生早期的表现,对烟雾的识别可实现林火早期预警。提出了一种基于烟雾静态特征和动态特征的识别算法。常见的烟雾静态特征的识别主要是在RGB色彩模式下进行烟雾静态特征识别,根据对烟雾像素特性的统计分析,提出在Lab色彩模式下进行,确定烟雾色彩检测模型。对于烟雾动态特征的识别,采用当前帧和背景图像进行差分的方法提取移动区域。在该算法设计中,首先计算自适应阈值,再通过自适应阈值生成自适应背景,有效解决了背景发生变化时误报的情况。对提取的移动区域再进行连通性判断,排除孤立点。以森林火灾视频集为对象,将所提出算法与文献方法进行了对比,结果表明,所提算法提高了检测率。
This paper intends to bring about an improved forest-fire identification algorithm based on the static and dynamic features of the smoke. Based on the static characteristics of the smoke,the proposed algorithm can first of all help to extract the suspected pixels, and, then, tries to make further judgment on the suspected pixel according to the dynamic features of the smoke, and, finally, enables the fire surveyors to decide the suspected pixel＇s connectivity. As is known, traditional fire-detecting systems used to apply the ion-optical smoking type or other physical or chemical means to detect where the fire comes out. Such methods are in many ways not feasible for the open forest fire detection since they may occur in a long distance or in a remote vast area. In addition, since the smoke is just an early indicator of the likely-to-be great forest fire, what we should do through identification is just to give out a warning signal of the fire as early as possible by identifying its indicator. Moreover, what we mean by the static characteristics of the smoke image mainly refers to color feature. The static feature identification of common smoke is mainly in RGB color mode, which identification can be carried out in the Lab color space according to the statistical analysis of the pixel features of the smoke. Therefore, we prefer to bring forward the study of color features which can be identified in the Lab condition, and establish the algorithm for identifying the smoke color. In our improved algorithm design, the identification can be done based on the analysis of the dynamic features of the smoke mainly extracted from the moving regions in the video by differentiating the background image. In order to solve the false positives due to the background change, we have first of all worked out the dynamic threshold according to the mean value and the standard deviation of the pixel image difference, and, next, try to reproduce the adaptive background via the above-said dynamic threshold, and last of all update the image for the background color identification. It is on this basis that it would be possible for the operator to perform the connectivity identification on the extracted mobile area and exclude the outlier so as to improve the robustness and identification rate of the image through the algorithm mode.
出处
《安全与环境学报》
CAS
CSCD
北大核心
2013年第2期210-214,共5页
Journal of Safety and Environment
基金
西南林业大学科学基金面上项目(2010MS07)
关键词
人工智能
森林火灾
烟雾颜色模型
动态特征
自适应背景
连通域分析
artificial intelligence
forest fires
smoke color model
dynamic characteristics
adaptive background
connected domain analysis
