Automatic Analog Pointer Instruments Reading Using Panel Detection and Landmark Localization Networks

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

1 Scopus citations

Abstract

Analog pointer instruments are widely used in industrial automation. However, reading their values by human is time and labour consuming. Most current pointer reading algorithms lack universality and can not handle complex and varying application scenarios. Therefore, this paper proposes a general pointer instrument reading scheme based on CNN. Our main contributions are two folds. First, we combine two sub-networks into one backbone, which has shortened the inference time. Second, we propose a light-weighted sampling decoder, which makes our scheme achieve a higher accuracy. Comparing to other algorithms, our scheme can deal with the case in which the instrument is too difficult to read, e.g., there is more than one similar pointer in one panel. Also, our scheme only needs less than 100 ms on a typical laptop for a 512 × 512 image. It is fast enough and can satisfy most requirements in industrial automation.

Original languageEnglish
Title of host publicationCognitive Systems and Signal Processing - 4th International Conference, ICCSIP 2018, Revised Selected Papers
EditorsFuchun Sun, Dewen Hu, Huaping Liu
PublisherSpringer Verlag
Pages3-14
Number of pages12
ISBN (Print)9789811379826
DOIs
StatePublished - 2019
Event4th International Conference on Cognitive Systems and Information Processing, ICCSIP 2018 - Beijing, China
Duration: 29 Nov 20181 Dec 2018

Publication series

NameCommunications in Computer and Information Science
Volume1005
ISSN (Print)1865-0929
ISSN (Electronic)1865-0937

Conference

Conference4th International Conference on Cognitive Systems and Information Processing, ICCSIP 2018
Country/TerritoryChina
CityBeijing
Period29/11/181/12/18

Keywords

  • Landmark localization
  • Panel detection
  • Pointer instrument reading

Fingerprint

Dive into the research topics of 'Automatic Analog Pointer Instruments Reading Using Panel Detection and Landmark Localization Networks'. Together they form a unique fingerprint.

Cite this