适应于户外场景下低成像质量的指针式表计读数智能识别方法  

Intelligent Recognition Method for Pointer-type Meter Reading Adapted to Low Image Quality in Outdoor Environments

作　　者：刘萌 王波[1] 罗鹏[1] 马富齐 张迎晨 王雷雄 LIU Meng;WANG Bo;LUO Peng;MA Fuqi;ZHANG Yingchen;WANG Leixiong(School of Electrical and Automation,Wuhan University,Wuhan 430072,China;School of Electrical Engineering,Xi’an University of Technology,Xi’an 710054,China)

机构地区：[1]武汉大学电气与自动化学院,武汉430072 [2]西安理工大学电气工程学院,西安710054

出　　处：《高电压技术》2024年第11期5034-5046,共13页High Voltage Engineering

基　　金：国家重点研发计划(2021YFB2401300);云南省科技厅科技专项(202202AD080004)。

摘　　要：由于户外场景下智能巡视表计设备采集图像时会受光照条件影响,设备采集图像质量较低,因而难以实现表计的准确读数。为此,该文首先提出一种户外场景下巡视表计的方法。该方法分为表盘检测与读数检测两部分,通过训练表盘特征高效检测模型进行表盘检测后,将检测到的候选框图像进行预处理以及基于轮廓检测的二值图重构,然后在上述二值图中依次进行起始刻度线检测、指针检测,最后使用角度法获得表计读数。经实验验证,该文方法可以克服户外场景下表计图像存在的图像模糊、反光以及存在阳光阴影等低成像质量的影响,在5种常见光照场景下的表盘目标平均误检率与平均漏检率分别为1.005%与0.505%,单张图像检测平均耗时28.1 ms,读数检测准确率为93.91%,对于复杂光照场景下的变电站指针式表计读数具有适用性与有效性。Due to the influence of lighting conditions on image acquisition by intelligent inspection meter devices in outdoor settings,the quality of captured images is often low,making it difficult to accurately read the meter.To address this issue,this paper introduces a method for outdoor inspection meter reading.The method comprises dial detection and reading detection components.After dials are efficiently detected by using a trained feature detection model,the detected candidate box images undergo preprocessing and binary image reconstruction based on contour detection.Subsequently,the scale line detection and pointer detection are performed in the reconstructed binary image,followed by obtaining meter readings using an angular method.Experimental verification demonstrates that the proposed method mitigates the effects of image blurring,glare,sunlight shadows,and other factors affect the image quality in outdoor scenarios.The average false detection rate and average miss detection rate for dial targets under five common lighting conditions are 1.005%and 0.505%,respectively.The average time for single image detection is 28.1 ms with a reading detection accuracy of 93.91%.This method proves to be applicable and effective for accurate reading of pointer-type meters at substations under complex lighting conditions.

关 键 词：表计读数识别 目标检测 概率霍夫变换 轮廓检测 深度学习 指针式仪表 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TM63[自动化与计算机技术—计算机科学与技术]