变电站指针式仪表检测与识别方法

发布时间：2018-03-13 12:47

本文选题：指针式仪表　切入点：检测与识别　出处：《仪器仪表学报》2017年11期 　论文类型：期刊论文

【摘要】：受限于复杂的电磁环境,变电站中的大量模拟式仪表需要人工读取示数,不利于变电站自动化管理。而目前针对仪表自动读数方法的研究大多基于预先获取到的高质量图像,其中仪表目标位于图像中央且占比较大,仪表表盘与相机平面平行,这需要大量预先的仪表测量与相机标定工作,不能满足实际电站环境下的使用要求。为解决上述问题,提出了一种完整的变电站指针式仪表的自动检测与识别方法。首先利用卷积神经网络模型检测当前视野下仪表目标的包围框位置,计算其距离视野中央的偏离值与图像占比,据此调整相机位置和缩放倍数。通过透视变换消除表盘平面与相机平面偏差造成的仪表图像畸变,通过霍夫变换检测仪表的表盘与指针,完成仪表读数识别。变电站实际测试实验结果表明,本方法最大读数误差仅为1.82%,对于复杂背景下多类别仪表的自动检测与识别任务具有良好的准确性与稳定性,可满足变电站实际应用需求。
[Abstract]:Limited by the complex electromagnetic environment, a large number of analog meters in the substation need manual reading, which is not conducive to substation automation management. The instrument target is located in the center of the image and occupies a large proportion, and the meter dial is parallel to the camera plane, which requires a large amount of pre-instrument measurement and camera calibration work, which cannot meet the requirements of the actual power station environment. In this paper, a complete automatic detection and recognition method for an exponential instrument in a substation is proposed. Firstly, a convolution neural network model is used to detect the position of the bounding frame of the instrument object in the current field of vision, and the deviation value from the center of the field of vision is calculated, and the ratio of the image to the image is calculated. The distortion of instrument image caused by the deviation between dial plane and camera plane is eliminated by perspective transformation, and the dial and pointer of instrument are detected by Hough transform. The actual test results of substation show that the maximum reading error of this method is only 1.82, which has good accuracy and stability for automatic detection and recognition of multi-class instruments in complex background. It can meet the practical application demand of substation.
【作者单位】：中国北方车辆研究所;
【分类号】：TM63;TP391.41

【相似文献】