Automatic reading of mechanical metering hardware in buildings using a low power edge device

Due to the greater demand for energy resources, monitoring is now necessary for their responsible and sustainable utilization. Collecting data from electricity meters is a crucial step towards achieving this. In most nations, this operation is completed manually at most once per month due to the limitations relating to cost and time. The average consumption for the preceding months is typically used to estimate and determine the current month's consumption. Due to increased billings that don't correspond to reality, this leads to numerous claims from customers. This project aims to develop an AI-based system that utilizes a cost-effective microcontroller to automate the data collection from electricity meters. The Convolutional Neural Network model developed using a manually constructed dataset serves as the centerpiece of the proposed system. The model's accuracy during the model testing phase was 97.63 %. Without the need for additional algorithms, this system can recognize full digits and intermediate digits (such as 4.5 and 5.5). The proposed system was tested and validated by experiments.