Electric energy meters play a crucial role in accurately measuring and recording the electricity consumption of residential, commercial, and industrial consumers. Traditionally, electric meters are read manually by a representative from the utility provider, who physically records and reads the meter to generate the bill. In recent years, advancements in energy measurement instruments have enabled the deployment of smart meters that offer detailed household-level energy consumption data. This study involves the development of an image processing and digit recognition-based Meter Reading System (MRS) designed to automate the process of electric meter readings. The system utilizes an ESP32-CAM to capture images of the electric meter and employs Convolutional Neural Network (CNN) algorithms to detect and recognize the digits on the meter. The recognized data is then stored and made accessible through an online web interface, providing users with real-time monitoring of their electricity consumption. The study demonstrates the successful implementation of various models for digit recognition, highlighting the importance of high-quality image capture and proper setup of the Region of Interest (ROI). Several tests were conducted to evaluate the performance of different models, with quantized models demonstrating significant efficiency by reducing processing time while maintaining accuracy. Among the models tested, DIGCLASS100Q proved to be the most accurate overall based on its recognition accuracy across the three different images. For future improvements, it is recommended to collect and train the system with digital images from local electric meters to enhance recognition accuracy.