林火监测中基于视觉的火焰检测方法

摘要

基于计算机视觉的火灾监测系统中对火焰的早期检测是防止森林火灾蔓延的关键.提出一种基于运动检测和像素颜色特征相结合的新的火焰检测方法.首先根据图像序列建立背景模型,突出显示运动部分;并采用邻域平均方法对目标和噪声进行分离.其次基于边缘特性,获取序列图像的边缘变化情况,将两者融合得到感兴趣(ROI)区域.最后,将亮度与饱和度分离,产生火焰的有效颜色空间.通过对火焰像素亮度(I)分量区域与饱和度(S)分量区域进行相关性分析来确定火焰.实验结果显示,方法具有良好的快速性,在较短的检测时间下,可以保证较高的检测率和较低的误检率,对火焰序列图像的检测率达到91.78％,对伪火焰图像的误检率达到7.11％.%Early detection of fire is the key to prevent the spread of forest fires in the fire monitoring system based on computer vision.This paper proposes a flame detection method based on motion detection and pixel color feature.Firstly,In this paper the background model is established according to the image sequence in order to highlight the motion part.And the neighborhood averaging method is used to separate the target and the noise,Secondly,based on the edge feature,the edge change of the sequence image is obtained.Combining that to obtain the target roi area.Finally,the brightness and saturation of the region of interest are separated,and effective color space of flame is produced.The flame is determined by analyzing the I region and the S region of the flame pixel.Experimental result show that the method proposed in this paper has good rapidity,the higher detection rate and lower false detection rate can be guaranteed in the short detection time.The detection rate of flame sequence image is 91.78％,and the false detection rate of pseudo flame image is 7.11％.