Text detection and recognition in raw image dataset of seven segment digital energy meter display

Karthick Kanagarathinam,Kavaskar Sekar

doi:10.1016/j.egyr.2019.07.004

Abstract

The article describes the collection of the dataset of raw images of digital energy meters display, text detection and recognition of seven segment numerals from collected samples that may be helpful in reducing the cost of advanced metering infrastructure (AMI). The presented dataset has tremendous potentials in fully automated optical character recognition (OCR) based electricity billing. The dataset has been named as ‘YUVA EB Dataset’ that has the collection of digital energy meter images. The images have been captured under day and night light conditions. The research work on recognizing the text from seven segment display in energy meters has been carried out using our dataset under the challenging text recognition conditions like tilted position, blurred, day and night light captured images. MSER and labeling method based OCR algorithm has been used for text detection and recognition.

Highlights

The advent of technology has charted an amazing and noble growth curve for the past two centuries
Various techniques using magnetic stripe, speech recognition, identification using radio frequency, bar code, and Optical Mark Recognition (OMR), and optical character recognition (OCR) fulfill the automation needs in various applications (Weigelt et al, 2010)
The objective of the article is to create raw image dataset and creating The Common Objects in Context (COCO) annotation JSON file using ‘Labelbox’ tool, to help the energy/water/gas consumption billing completely free from the task of the meter readers by implementing OCR to extract the value from digital meters and to implement maximally stable extremal regions (MSER) based preprocessing to overcome the challenges in text detection and text recognition in camera captured images

Summary

Introduction

The advent of technology has charted an amazing and noble growth curve for the past two centuries. The advancement comes to a great distance from punch cards in the 1950s to the graphical interface in the past decade to human-based interactions in the present and the future. The usage of mouse and the keyboard is comfortable for last few years to assist as interfaces between human and the computer. The ability to use human-based computer interaction would make things easier mechanically for the users, but would be difficult to succeed for the researchers. The pioneering achievements due to continuous research in man–machine communication may bring the scenario like interactions between human. Various techniques using magnetic stripe, speech recognition, identification using radio frequency, bar code, and Optical Mark Recognition (OMR), and OCR fulfill the automation needs in various applications (Weigelt et al, 2010)

Objectives

Results

Conclusion