视频监控系统中火灾检测技术的研究

【作者】 张志英；

【导师】 康波；

【作者基本信息】 电子科技大学 ， 模式识别与智能系统， 2010， 硕士

【摘要】 随着城市化进程的不断加深,高层住宅、地铁、体育馆、学校、商场、公园等越来越多的出现在人们周围,与此同时火灾对城市造成的损失也越来越大。传统的火灾监控技术已不适应社会的快速发展。近年来,随着计算机和数字图像处理技术的快速发展,一种新的基于视频图像检测的安全监控系统逐渐成为主流的研究方向。本文主要研究的就是视频监控系统中的火焰检测技术并实现了一套火焰检测的具体方法,它通过分析处理从摄像机中获得的现场视频图像,以达到对火灾实时的自动检测和监控。本文首先讨论了运动前景目标的检测方法,这是火焰检测的基础,直接影响火焰检测的结果和效率。根据实际复杂场景的需要,系统采用改进后的码本算法进行运动前景目标的提取,与传统的混合高斯建模法相比,此检测方法更加准确,速度更快,能处理更多动态背景。通过运动前景的检测后,包括火焰在内的可疑运动区域被分割了出来。要实现火焰的检测,就要将火焰区别于其他物体的特征提取出来。火焰在燃烧过程中具有复杂的变化过程和表现形式,通过观察,本文提取出火焰的颜色特征、闪动频率特征、相对稳定性特征、空间信息丰富及形状不规则特征。首先,建立颜色模型,如果可疑的运动区域符合此模型要求,则认为其具有火焰的颜色。针对闪动频率特征,本文提出利用短时过零率和小波时域分析两种方法检测,均取得了不错效果。与此同时采用短时轮廓重合率描述可疑区域的相对稳定性,利用空间小波变换和圆形度分别度量可疑区域的空间变化信息和形状不规则性。最后,为了提高识别的准确度,使用加权投票机制将多种特征有效融合综合判定火焰的存在。在对不同环境中的大量视频进行测试后,实验结果表明,本文提出的火焰识别方法错误率低,且具有较好的实时性和一定的抗干扰能力,能广泛的应用于城市中多领域的安全消防系统中。更多还原

【Abstract】 With the deepening of urbanization, there are more and more skyscraper, subway, gymnasium, schools, shopping malls and parks emerging around people. At the same time, the losses that caused by fire is also growing in the city. The traditional fire monitoring methods have been not suited to the rapid development of the society. In recent years, with the rapid development of computer and digital image processing technology, a new safety monitoring system based on video image detection has gradually become the mainstream of research.The thesis mainly studies the flame detection technology in the video surveillance system and realizes a flame detection method which achieves real-time automatic fire detection by analyzing and processing live video images that captures from camera. Firstly, the method of moving region detection which is the basis of flame detection and directly affects the flame detection results and efficiency is discussed. According to the needs of the actual complex scene, the system uses the improved CodeBook algorithm to extract the moving regions. Compared with the methods of mixture of Gaussians, The Codebook method can detect more accurate, faster and handle more dynamic background. After moving regions detection, the suspicious moving areas are extracted.To achieve the flame detection, we must extract the characteristics which can distinguish other objects. The flame has complex change process and forms in the burning process. Its color features, flicker frequency characteristic, relative stability characteristic, space change information and irregular shape characteristic are extracted by observing. Firstly, the color model is established. If the suspicious region meets the requirement of the color model, we think that it is a fire-colored region. And then two methods: short-time zero crossing rate and temporal wavelet transform are proposed to detect flicker frequency. At the same time we use short-time contour coincidence rate to describe relative stability of the suspicious region. Spatial wavelet transform and circularity are utilized to analyze space change information and irregular shape respectively. At last, all of the characters extracted above are combined by weighted voting scheme to achieve accurate flame detection and judgments. After testing many videos in different environments, experiments show that this method can detect the flame effectively, and has better real-time and anti-interference ability. It can be widely applied in many domains of the city fire safety system.更多还原