【作者】 张娜；

【导师】 吴成东； 贾子熙；

【作者基本信息】 东北大学 ， 控制工程（专业学位）， 2011， 硕士

【摘要】 传统的自来水公司需要专门人员到用户家中进行人工抄表和收费管理,既耗费了人力又给用户带来麻烦。目前出现很多新型水表便于自动读取水量和收费。但由于居民及工业中普遍已安装的水表大部分是多指针型仪表,更换新水表,涉及管道和电子线路的改造问题,既费时费力难以推广也需要很大的成本。所以通过水表上方安装摄像头,远程传送水表图片,经数字图像处理技术实现水表读数的自动识读具有重大意义。本文设计了多指针型水表自动识读系统。论文的主要研究内容如下：(1)针对表盘灰度图像中指针的灰度值与表盘背景灰度值相似的问题,本文采用多尺度Retinex低通滤波算法对水表彩色图像进行光照补偿处理与图像增强。使原图中红色指针的对比度更加明显,进而为提高指针区域分割的准确性提供了可靠的数据基础。(2)选用C均值聚类算法对水表指针区域的进行图像分割。本算法充分考虑了红色指针分量的灰度值与表盘背景和黑色指针灰度值之间存在的分类特性,根据图像灰度级将像素分为三类,并通过C均值聚类算实现分类。仿真结果表明该方法优于大津法、区域生长法等常规分割算法。(3)在面向二值图像的图像后处理阶段,本文首先基于二值图像形态学操作进行去噪处理。对于个别无法通过图像相态学操作进行滤除的噪声区域,本文提出根据质心距离曲线的分类识别算法对指针区域和非指针区域进行区分,达到全面去噪的目的。(4)应用最小二乘法实现指针圆心坐标及表盘水平零刻度线方向的求取,并通过相应的几何关系进行指针读数的快速计算。设计了对于多指针水表读数获取的自动识别算法,该算法可分别对子表盘数量级以及各子表盘指针刻度进行自动识读,经实验表明该算法具有较好的鲁棒性与识别准确度。更多还原

【Abstract】 For the majority of tap water companies, the record and charge management of the water consumption are all to depend on the manual methods. The automatic recognition of water meter is one of the important process steps in automatic meter reading. It is also a key of integrated into the intelligent system. To improve the present backward condition, an automatic reading system of water meter has been developed so that water meter can be automatically read by applying image processing technique, which is completely new in our country at present.This thesis includes the following related parts:First, the feature of water meter image is analyzed. Light compensation was used to improve the brightness of the image. The algorithm of Light compensation based on Retinex algorithm is researched for Automatic Reading system. After Light compensation the RGB sub-ratio of the image was changed. The contrast of the red pointer and the background of the water meter was greatly increased. And to extract the part of the pointers in the special space of the water meter is better.Second, pointer extraction based on Fuzzy C-Means Clustering (FCM) was presented for processing non-uniform illumination in image segmentation. Considering the gray value of the red pointer is between the gray of the pointer dial background and black pointer, the gray image eventually divided into three categories.Third, mathematical morphological filtering and matching based on centroid distance increment matrix were applied for eliminating the noise from the pointer region. The size of some individual noise area and the pointer region area is closed, so morphological denoising method is not enough. The algorithm of matching based on centroid distance increment matrix is better in eliminating the noise of the water meter image.Four, the center and the reference orientation of the water meter was determined by using the least-square method. The reference orientation of the water meter vector direction is the key to the whole student algorithm. Then, the Automatic reading algorithm for water meter was studied.更多还原