Make your analog meter snitch on itself for under €10

What it does

AI-on-the-edge-device is firmware for an ESP32-CAM that periodically photographs analog utility meters—water, gas, or electricity—and extracts the current reading using on-device TensorFlow Lite inference. It handles the full computer-vision pipeline internally, from image alignment and ROI extraction to digit recognition, then pushes the result to an MQTT broker, InfluxDB, or a REST endpoint.

The interesting bit

Instead of uploading photos to the cloud for processing, all inference happens on the microcontroller itself. The project squeezes an entire CV pipeline—feature detection, alignment, illumination, and TFLite digit recognition—onto hardware that costs less than €10 and fits in a 3 cm × 4.5 cm × 2 cm case.

Key highlights

• Runs entirely on the ESP32; no cloud dependency for image recognition
• Ships with a web interface for administration and over-the-air firmware updates
• Integrates with Home Assistant, InfluxDB v1/v2, MQTT, and a REST API
• Includes 3D-printable housings tailored for water, power, and gas meters
• Requires only an ESP32 board with a compatible camera and an SD card

Caveats

• Developer capacity is limited, so several feature requests in the backlog are not currently being pursued
• Initial firmware installation requires a USB connection, and the docs note you’ll need “some practical skills”

Verdict

Home-automation enthusiasts stuck with analog meters should take a look; anyone expecting a polished consumer gadget or cloud-native SaaS will be disappointed.