Reading electrical energy consumption in analog meters using artificial vision services in the cloud

Abstract

Currently, in Peru, the manual process of electricity consumption readings carried out house by house is still maintained. In this process there are errors in the reading, which leads to erroneous billing of electrical consumption. Given this, an alternative reading solution is proposed based on images and using computer vision or artificial vision technology. For this purpose, a set of images was generated with consumption readings from an analog meter in the town of Barranca, which, the images, as they were taken, went through an automated mechanism that each of the vision services used. computer from top 3 cloud providers Azure, Amazon, Google. The study aims to determine which computer vision technology has the greatest precision in determining the value of electrical consumption based on the image of an analog meter. The computer vision services made available by each cloud provider were used, for Azure the Azure AI Vision service was consumed, for Amazon the Amazon Rekognition service was consumed and for Google the AI Vision service was consumed. 100 images of meters with their electrical consumption were collected, which served as the data set of this investigation. For the solution proposal, a model with 6 stages was implemented: Collect Images, Identify the Electrical Consumption Region, Process Electrical Consumption Detection with Azure Computer Vision, Process Electrical Consumption Detection with Amazon Rekognition, Process Electrical Consumption Detection with Google Vision AI and Evaluate Results. The implemented model obtained the results that the cloud provider Amazon with its computer vision service Amazon Rekognition had the highest precision in detecting the value of electricity consumption.