jomjol/AI-on-the-edge-device

AI‑on‑the‑Edge‑Device – Digitizing Analog Meters with an ESP32‑CAM

What it is – A small, low‑cost hardware‑software kit that runs TensorFlow Lite inference on an ESP32‑CAM to read analog utility meters (water, gas, electricity). The device captures a picture of the dial, extracts the region of interest, runs a tiny neural network to recognise the digits, and then makes the reading available via MQTT, InfluxDB, a REST API or a Home Assistant integration.

Key components

• ESP32‑CAM (≈ 3 × 4.5 × 2 cm, < 10 €) with built‑in camera and LED illumination.
• TensorFlow Lite wrapper for on‑device inference.
• Image‑pre‑processing (feature detection, alignment, ROI extraction) performed on the ESP32.
• Web UI for configuration, OTA firmware updates and live preview.
• Connectivity: Wi‑Fi, MQTT, InfluxDB 1/2, REST API, Home Assistant auto‑discovery.

How it works

1. The firmware flashes the ESP32‑CAM.
2. At configurable intervals the camera snaps a photo of the meter.
3. The image is cropped to the dial area, normalised and fed to a TFLite model that outputs the numeric reading.
4. The result is pushed to the chosen back‑ends (MQTT broker, InfluxDB, etc.) or served via the built‑in REST endpoint.

Typical use‑cases

• Home‑automation: automatically log water, gas or electricity consumption.
• Retro‑fit legacy meters without digital output.
• Low‑power, offline‑first monitoring where cloud processing is undesirable.

Getting started

1. Hardware – Any ESP32 board with a supported camera module (see the hardware‑compatibility page). Optionally 3‑D‑print a case from the provided Thingiverse links.
2. Download – Grab the latest release from the GitHub Releases page.
3. Flashing – Use the browser‑based Web Installer (or Espressif’s Flash Tool / ESPTool) to flash the firmware via USB.
4. SD‑card setup – After the firmware boots, the device creates a Wi‑Fi access point; connect to it and let the built‑in script format and populate the SD card, or manually copy the config files.
5. Configuration – Access the web interface to set Wi‑Fi credentials, MQTT/InfluxDB endpoints, capture interval, and OTA update settings.
6. Integration – Enable Home Assistant discovery or point your MQTT client / InfluxDB to the device’s topics.

Documentation & support

• Full docs (installation, hardware compatibility, build instructions) are hosted at https://jomjol.github.io/AI-on-the-edge-device-docs/.
• Community help via the GitHub Discussions page; bugs and feature requests go to Issues.
• A collection of YouTube tutorials demonstrates the build and deployment steps.

Why it matters It showcases practical edge‑AI: a microcontroller runs a neural network locally, eliminating the need for constant cloud connectivity while keeping costs low. This makes AI‑enabled sensing accessible for hobbyists and small‑scale IoT deployments.

---

Quick links

• Source code: code/ (C++ firmware)
• Web installer: https://jomjol.github.io/AI-on-the-edge-device/
• 3‑D printable cases: Water, Power, Gas meters on Thingiverse/Printables
• Donate: PayPal link for supporting the developer