According to the requirement of power equipment automatic inspection,we present an intelligent reading recognition method for pointer meters based on computer vision technology.The dial in a test image is located as the interested region by using the color histo-gram distribution feature and brightness gradient feature.Illumination equalization is processed on the region to eliminate the effect of shadow.The directional binary descriptor ORB algorithm is used to calculate the perspective transformation matrix from the template im-age to the test image,then the matrix is adopted to make a tilt correction on the dial region.An adaptive threshold method is adopted to transfer the dial region to a binary image and the centripetalism of dial gauge lines is used to calculate the pointer rotation center.Then an improved Hough algorithm is utilized to detect the pointers,and the readings are calculated with the deflection angles of the pointers ac-cording to the layout of the dial.The experiments show that this method can address several kinds of dial layouts and make the recognition in different view angles.Its recognition error is under 2.5%,showing that it is very effective and adaptable,achieving real-time perform-ance and high accuracy.%针对电力设备自动化巡检的需求,提出了一种基于计算机视觉技术的电力指针式仪表读数智能识别方法.该方法利用表盘的颜色直方图分布特征和亮度梯度特征在整幅图像中定位出表盘作为感兴趣区域,通过光照均衡处理消除阴影影响,应用ORB算法计算待测图像与模板图像之间的透视变换矩阵,并对表盘区域进行倾斜校正;采用自适应阈值进行表盘区域的二值化处理,通过表盘刻度线向心性计算得到指针旋转中心,并由改进的霍夫直线检测算法进行指针定位检测,通过指针偏转角度及表盘格局计算获得相应读数.实验结果表明,该方法能适应多种类型表盘格局和不同视角下指针读数识别,其识别精度和准确率高、适应性强,实时性好,相对误差小于2.5%.
展开▼