一种基于质心距脉动特征的火焰轮廓识别算法  

Method of identifying fire flames in vision based on the variation contours

作　　者：沈诗林[1] 陈礼涛[1,2] 袁非牛[1,3] 于春雨[1] 乔利锋[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230027 [2]江苏省淮安市公安消防支队,江苏淮安223001 [3]江西财经大学信息学院,南昌330013

出　　处：《安全与环境学报》2009年第1期135-138,共4页Journal of Safety and Environment

基　　金：安徽省科技攻关计划项目(08010302203)

摘　　要：火灾视觉特征的提取是火灾图像探测的关键问题。提出了一种描述火灾火焰轮廓脉动信息的火焰图像质心距脉动模型,即先用质心距来描述火焰的轮廓信息,再利用傅立叶描述子描述火焰轮廓变化的时空信息,采用距离来度量相邻两帧图像轮廓的变化幅度。实验结果表明,对于火灾火焰视频图像而言,其相邻两帧之间的脉动幅度、变化频率要明显高于受控燃烧、光源干扰等非火灾情况,可有效应用于火灾火焰的识别。研究表明,该方法能有效地探测火灾的发生,降低系统的误报率。This paper is aimed at introducing a method of identifying fire flames in vision based on the variation contours. As is known, conventional point smoke and fire detectors are widely used at present fire-fighting practice to detect the presence of certain particles generated by smoke and fire by identifying their ionization and photometry. However, a big weakness of point detectors lies in that in large rooms, it may take a long time for smoke particles to reach a detector and, moreover, it is impossible to operate then in an open space. On the other hand, the vision based fire detection is a good method for large or open spaces and in turn the research in this way has become very active recently. ＂However, the method is still facing great technical challenges since its high false alarming rate. A key problem in this way is not easy to pick up vision features. In addition, the flickering of fire flame is also a common phenomenon in favorite for detecting the fire plume and air flow. Since the contour of fire flame keeps changing with the time, it is necessary to use a CCD camera to gather the video images of fire flame. In doing so, the following methods are supposed to pre-process the image. Firstly, median-filtering system is used to remove the noise of the image. Secondly, pre-mixed background and difference models are used to detect the moving region in the video image. After that, it would be easier to pick up the contour of the image moving region. And, finally, we have brought up forward a flickering model of centroid distance of the contour of fire flame image to define the characteristics of the image, with the distance model of flickering contours between the two successive frames presented in the normalized Fourier descriptors. The results of our experiment show that the algorithm is effective for fire recognition and therefore significant for improving the accuracy of the true fire alarm and reducing the those of the false ones in the meantime time.

关 键 词：安全工程 视觉火灾探测 傅立叶描述子 质心距脉动模型 火焰轮廓 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] X932[自动化与计算机技术—计算机科学与技术]