节点文献

基于图像信息融合技术的火灾识别算法研究

Study of Fire Detection Based on Image Date Fusion Technology

【作者】 高娜

【导师】 董爱华;

【作者基本信息】 河南理工大学 , 控制理论与控制工程, 2009, 硕士

【摘要】 高层建筑火灾具有传播速度快、灭火和营救都比较困难等特点,因此其有效防范及早期预测十分重要。针对传统火灾探测技术存在的不稳定、误判率高等缺点,本文着重分析了室内火灾图像与常见干扰光源图像的特点,并在此基础上研究了火焰图像采集、去噪、分割、边缘提取的算法,提出用火焰区域面积变化、质心位置、尖角数、圆形度等信息作为火灾判据,并用模糊神经网络对以上特征参数进行数据融合,作出火灾判断。本文采集了大量火灾试验图片和常见干扰光源的图片用于实验分析;通过对不同的图像的分析,确定对动态图像序列采用差影法获取可疑火焰区域图像;用中值滤波法去除图像噪声;用基于灰度直方图的中间阈值法进行图像分割,并在此基础上将图像二值化。以相邻帧图像中可疑火焰区域的面积变化、质心位置随机性变化以及单幅图像中可疑火焰区域的尖角数和圆形度等形态特征对火灾进行判别。在彩色图像分割算法的研究中,本文分析研究了现有的图像分割方法,针对最大类间方差法(Otsu法)易受图像直方图形状的限制以及分割效果不稳定等问题,提出了一种火焰图像分割的综合方法:利用HSI颜色空间参数对火焰图像建立彩色空间模型,根据火焰图像的S(饱和度)、H (颜色)等特征参数对图像进行预处理以确定火焰疑似区域,再用最大类间方差法进行阈值分割。通过实例仿真得到了轮廓清晰的分割图像。试验结果表明,该方法能够弥补最大类间方差法的局限性,扩展该方法的适用范围。最后,论文给出了模糊神经网络的具体结构和输入输出单元的设计方案,并使用大量的火灾图像样本和干扰模式的图像进行了验证。试验结果表明,基于模糊神经网络的信息融合算法能够有效地识别出火灾火焰,并且具有良好的抗干扰能力。

【Abstract】 For the fire’s characteristics of spreading quickly in highrise and difficulty of fighting and rscue,it is very important to detect and guard effectively in early stage. The shortcoming is also studied about conventional fire detection systems. The characteristics of fire image and familiar Interference light source have been analyzed in this dissertation, and the arithmetic has been studied about flame image acquisition, denoising, segmentation and edge extracting. The topic introduction four criterions: the area increases gradually, the centroid changes random, the taper angles, and the circularity. The Fuzzy neural network(NN) is used in Information Fusion to decide the fire.The massive images of fire and familiar interference light source are gathered for test analysis. By analyzing different images, method of difference image is used to obtain suspicious flame region image, median value filter method is used to remove the noise of image and intermediate threshold is used to segment images and implement binarization images. The fire flame is distinguished by the morphological characters of the area change and random centroid position of suspicious flame region which are from adjacent images, as well as the circularity and taper angle of suspicious flame region in each image.After analyzing image segmentation methods now available, to solve gray-level histogram shape limit and unsteady segmentation effect of the Otsu method, the paper provides a method of flame image segmentation which uses the Otsu method combined with the HSI color-space parameters. First build a color space model of flame image based on HSI color space parameter to preprocess the flame image using Saturation(S-parameters)、hue(H parameters) of image features, then segment the flame image according to the threshold value with the Otsu. At last, a clear profile of segmented image can be obtained by simulation. The test results show that this method can make up some shortcomings of the Otsu and widened its application range .The detail structure of the Fuzzy NN and the detail design precept of input and output layer have been given. The experiment results show that the BPNN fire detection algorithm can recognize the fire flame effectively and has good anti-interference capability.

节点文献中: 

本文链接的文献网络图示:

本文的引文网络