两种火灾烟雾识别方法的研究

(西安科技大学 计算机科学与技术学院,陕西 西安 710054)

火灾预警; 烟雾检测; 数字图像处理; 颜色特征; 小波变换

Two fire smoke identification methods
LI Na,QI Ai-ling,JIA Peng-tao,GONG Shang-fu

(College of Computer Science and Engineering,Xi'an University of Science and Technology,Xi'an 710054,China)

fire early warning; smoke detection; digital image processing; color characteristics; wavelet transform

备注

接触式火灾探测器在大空间和室外场景存在局限性,结合视频监控系统和数字图像处理技术,进行探测和预警成为火灾探测领域里的新研究方向。针对不同场景下烟雾识别算法的选择问题,研究颜色统计特征和小波变换两种方法在烟雾图像识别中的应用,分析算法的具体过程及其适用性。火灾烟雾图像包含丰富的颜色、纹理等特征参数,将烟雾图像转换为二值图像,采用数字图像处理的方法对烟雾的RGB颜色特征进行统计,分析其颜色距离与人类视觉的关系,提取图像中的烟雾点特征,识别是否有火灾烟雾生成以达到预警的目的 。小波变换方法利用烟雾背景图像纹理模糊即信号衰弱的特点,对目标前景进行提取。利用连续目标图像计算均值背景,通过与均值背景的对比,捕捉图像信号衰减的连通区域,获取烟雾图像。通过实验,从图像要求、烟雾对象、时间复杂度、适合场所几方面得出颜色统计特征和小波变换方法在火灾烟雾识别中的性能比较。结果 对不同场景下烟雾图像识别方法的选择有指导作用。

Contact fire detectors have limitations in large space and outdoor scenes,so combining video surveillance system with digital image processing technology to conduct fire detection and early warning has become a new research direction in the field of fire detection.Aiming at the selection of smoke recognition algorithms in different scenarios,this paper studies the application of color statistical features and wavelet transform in smoke image recognition,and analyzes its applicability through the specific process of the algorithm.Fire smoke image contains rich color,texture and other characteristic parameters.In the statistical method of color feature,smoke image is converted into binary image,and the RGB color feature of smoke is counted by digital image processing method.Then,the relationship between color distance and human vision is analyzed to extract smoke feature,and then the purpose of identifying whether there is fire smoke and early warning is achieved.Texture of fume background image is blurring,which is regarded as signal weakening,and this is used to extract target foreground in the method of wavelet transform.Mean background is calculated by using continuous target image.By comparing with mean background,the connected area of image signal attenuation is captured and smoke image is obtained.Through experiments,the performance indicators of color feature statistics and wavelet transform methods in fire smoke recognition are obtained from image requirements,smoke objects,time complexity and suitable places.The results will guide the selection of smoke image recognition methods in different scenarios.