AI-on-the-edge-device

This project lets you read your home's physical utility meters, such as water, gas, and electricity meters, automatically using a cheap camera-equipped microcontroller. The idea is to attach a small device near the meter's display, take photos of it at regular intervals, and use image recognition software running directly on the device to extract the numeric reading. That reading is then sent over your home network to other software so you can track consumption over time.

The hardware involved is an ESP32-CAM, a small circuit board roughly the size of a postage stamp that costs under ten euros and includes a Wi-Fi chip, a microcontroller, and a camera. The software running on it uses TensorFlow Lite, a version of Google's machine learning library designed for low-power devices, to recognize the digits on the meter face. The device handles all image processing locally without sending photos to any cloud service.

Once the device reads a meter value, it can deliver that number in several ways: via MQTT (a standard messaging protocol used in home automation), to an InfluxDB time-series database for logging, or through a REST API that other software can query. There is also built-in integration with Home Assistant, a popular open-source home automation platform.

Setup involves flashing firmware to the ESP32 board (using a browser-based installer or a command-line tool), preparing a micro SD card with configuration files, and mounting the device near the meter. A web interface running on the device lets you configure which part of the meter image to read and monitor what the AI is recognizing. Firmware updates can be done wirelessly after initial setup.

The README includes links to 3D-printable housings for water, power, and gas meters, along with YouTube tutorials from community members who have installed it. Full documentation is hosted separately on a dedicated documentation site.