What is it about?
This article presents the conception of a new method developed mainly in Python to automate the reading process of water meters with an analog display using computer vision and machine learning. A camera captures the consumption value in the water meter, and the yielded image undergoes image processing until the digits are detected and isolated. Then the digits are passed into an SVM machine-learning model that carries out a high accuracy OCR. The software is executed over an ARM platform running Linux. The data resultant from the automated metering, such as the device identification number, event date and time in UTC, consumption value, volume and time variations, flow, and display image, are locally stored and transmitted to a cloud server through VPN in a Wi-Fi and cellular network connection, or by SMS, enabling a remote supervision. Thereby, the automatic metering method features a new way to perform predictive analysis and management of water and meters proactively and can be replicated for digital-display water meters, as well as extended to handle automatic metering on electricity and gas meters as well.
Featured Image
Read the Original
This page is a summary of: Deployment framework for the Internet of water meters using computer vision on ARM platform, Journal of Ambient Intelligence and Smart Environments, January 2020, IOS Press,
DOI: 10.3233/ais-200544.
You can read the full text:
Resources
Water meter automated metering, Demo
It shows how the setup of the computer-vision machine is performed on the software created. The software runs on an Arch Linux operating in a Raspberry Pi 2 Model B.
Water meter automated metering, A Quick Overview
The video demonstrates the integration between a test bench that was built and the software developed. The program in execution updates metering data of interest every 20 seconds, at least, and performs a set of routines such as ROI localization, digits detection, and digits recognition, which in turn is achieved by means of Support Vector Machine (SVM). k-Nearest Neighbor and Multi-Layer Perceptron were also trained and cross-validated. The three mentioned machine-learning models were generated with a data set of 5,000 digits snippets (500 of each digit).
Water meter automated metering, Machine zeroing
It shows how the setup of the computer-vision machine is performed on the software created. The software runs on an Arch Linux operating in a Raspberry Pi 2 Model B.
Web page for metering monitoring
Online interface for metering supervision.
Contributors
The following have contributed to this page
