Laser & Optoelectronics Progress, Volume. 59, Issue 24, 2410001(2022)

Fully Automatic Reading Recognition for Pointer Meters Based on Lightweight Image Semantic Segmentation Model

Fuhai Yan1,2, Wangming Xu1,2,3、*, Qiugan Huang1,2, and Shiqian Wu1,2
Author Affiliations
  • 1School of Information Science and Engineering, Wuhan University of Science and Technology, Wuhan 430081, Hubei, China
  • 2Institute of Robotics and Intelligent Systems, Wuhan University of Science and Technology, Wuhan 430081, Hubei, China
  • 3Engineering Research Center for Metallurgical Automation and Detecting Technology of Ministry of Education, Wuhan University of Science and Technology, Wuhan 430081, Hubei, China
  • show less
    Figures & Tables(13)
    Flow chart of proposed method
    Network structure of modified image semantic segmentation model
    Gaussian heat map label generation
    Sematic segmentation results and separated binary images. (a) Semantic segmentation results; (b) binary image of scale lines; (c) binary image of pointer; (d) binary image of scale-range numbers
    Correction for skew and distorted image. (a) Dial ellipse fitting result; (b) schematic graph of perspective transformation
    Image correction and denoising results. (a) Scale line; (b) pointer; (c) scale-range numbers
    Image polar transform. (a) Polar coordinate transform coordinate system; (b) polar coordinate transform result
    Location and repair of scale lines and pointer. (a) Contour refinement result; (b) center line positioning result; (c) scale line repair result
    Comparison of semantic segmentation effects for different lightweight models
    Gray images (left) and binary images (right) of scale-range numbers
    • Table 1. Result of ablation experiments

      View table

      Table 1. Result of ablation experiments

      ModelmIoUPA
      CGNet70.9497.69
      Contrast Model A71.1697.69
      Contrast Model B70.3297.65
      Contrast Model C76.5698.27
      Proposed model77.3798.29
    • Table 2. Comparison between different lightweight sematic segmentation models

      View table

      Table 2. Comparison between different lightweight sematic segmentation models

      ModelModel size /MBmIoU /%PA /%Reasoning speed /(frame·s-1
      ICNet10854.8096.9849
      BiseNet48.973.4397.82219
      Fast-SCNN4.5368.5997.48217
      CGNet2.0970.9497.6994
      Proposed model2.7077.3798.2927
    • Table 3. Meter reading results of typical test images

      View table

      Table 3. Meter reading results of typical test images

      No.Scale rangeManual readingReading of proposed MethodRelative Error /%
      10.60.030.030.00
      21.61.101.090.63
      31.60.610.600.63
      42.50.650.650.00
      51.60.630.620.63
      60.60.030.030.00
      72.50.660.660.00
      81.61.091.071.25
      92.50.300.310.40
      101.60.620.610.63
    Tools

    Get Citation

    Copy Citation Text

    Fuhai Yan, Wangming Xu, Qiugan Huang, Shiqian Wu. Fully Automatic Reading Recognition for Pointer Meters Based on Lightweight Image Semantic Segmentation Model[J]. Laser & Optoelectronics Progress, 2022, 59(24): 2410001

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Image Processing

    Received: Sep. 1, 2021

    Accepted: Oct. 27, 2021

    Published Online: Jan. 11, 2023

    The Author Email: Xu Wangming (xuwangming@wust.edu.cn)

    DOI:10.3788/LOP202259.2410001

    Topics