基于视频动态纹理的森林火灾检测算法.doc

中图法分类号: O29: TP391.4 文献标识码: A 文章编号: 1006-8961(2012) 基于视频动态纹理的火灾检测 邵婧1，2 ， 王冠香2，3 ， 郭蔚1，2 1.河北工业大学理学院, 天津 300401； 2.北京国际数学研究中心, 北京 100871；3.北京大学数学科学学院, 北京 100871 摘 要 ：长期以来，森林火灾检测一直都是世界范围内的一个重要研究课题，对于保护地球环境及人类安全都有重要意义。基于视频监控的火灾检测对于实时性和正确性要求很高，而传感式探测器和传统图像型探测器不能满足要求。本文提出了一种基于动态纹理特征分析的新型图像型火灾检测算法，对于森林这样的复杂大空间场景尤为适用。通过对CCD摄像机拍摄的视频图像，建立线性动力系统（Linear Dynamic System, LDS）模型，分析其动态纹理特征，最后利用Adaboost分类器判断火灾是否存在。实验结果表明，此算法能够达到95%的检测准确率，且具有较好的应用前景。 关键词 ：森林火灾检测；动态纹理；线性动力系统；马氏距离；Adaboost分类器 Fire detection based on video dynamic texture Shao Jing1, 2, Wang Guanxiang2, 3, Guo Wei1, 2 1. College of Science, Hebei University of Technology, Tianjin, 300401; 2. Beijing International Center of Mathematical Research, Beijing, 100871; 3. College of Mathematical Science, Peking University, Beijing, 100871 Abstract: For a long time, forest fire detection is one of the most important research areas all over the world. It is crucial for protecting the earth and human survival. Real-time and accuracy is the essential need in surveillance system, especially in forest fire detection. However, the traditional sensor-based and conventional image-based fire detection approaches can not satisfy the needs. To achieve an effective detection of such fire, an image-based forest fire detection algorithm based on dynamic texture analysis is proposed. The first step is to establish Linear Dynamic System (LDS) model on the video image sequences captured by CCD camera, then analysis the dynamic textures, and finally determine the existence of the fire by Adaboost classifier. Experimental results demonstrate that the rate of accuracy of the proposed algorithm can achieve 95%. It is efficient and robust with good application prospects. Key words: forest fire detection; dynamic texture; linear dynamic system; martin distance; Adaboost 0 引 言 及时准确的火灾检测是控制火灾蔓延的一个重要手段，也是监控系统中尤为重要的一个环节。已有的火灾检测方法大体分为两类：基于传感器原理设计的火灾探测器和基于图像处理的火灾检测系统。传统的火灾探测器是通过对火灾参量（如温度、烟雾、光、燃烧气体等）的传感分析，在某种程度上判定是否有火灾的发生。然而在一些大空间场合以及环