← Back to Portfolio
Energy / Utilities

Automated Meter Reading with Computer Vision

A production-grade AI system that reads electrical meter indices from photos with high accuracy, deployed as a mobile-compatible REST API.

The Challenge

Manual meter reading is error-prone and expensive to scale. The client needed a solution that field technicians could use from a standard Android phone to capture and submit meter readings automatically, eliminating transcription errors.

Our Solution

We developed a custom YOLOv5 object detection model trained on hundreds of annotated meter images. The model detects and classifies digits (0–9) plus decimal separators with high precision. It was converted to TensorFlow Lite format for efficient inference on Android devices and wrapped in a Flask REST API with Swagger documentation and HTTP authentication.

The Result

The system achieved high accuracy on the test set, successfully deployed to production, and integrated with the client's mobile workflow. Field technicians can now submit readings by simply photographing the meter.

Technologies Used

PythonYOLOv5TensorFlow LiteFlaskOpenCVSwagger

Have a Similar Project?

Let's discuss how we can help.

Get In Touch