高电压技术2024，Vol.50Issue(11)：5034-5046,13.DOI:10.13336/j.1003-6520.hve.20231408
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
摘要
Abstract
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 me-ter 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 av-erage 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 accu-racy 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Key words
meter reading identification/object detection/probabilistic Hough transform/contour detection/deep learn-ing/pointer instrument引用本文复制引用
刘萌,王波,罗鹏,马富齐,张迎晨,王雷雄..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[J].高电压技术,2024,50(11):5034-5046,13.基金项目
国家重点研发计划(2021YFB2401300) （2021YFB2401300）
云南省科技厅科技专项(202202AD080004).Project supported by National Key R&D Program of China(2021YFB2401300),Science and Technology Department Project of Yunnan Province(202202AD080004). （202202AD080004）
