[Custom Component] AI Meter Reader and general tflite esphome implementation + Call for Contributors!

Share your Projects!
,
1

Hi everyone! :wave:

I’m excited to share a major update to the ESPHome AI Component. This project lets you turn an ESP32 camera into a fully offline, privacy-first meter reader (water, gas, electricity, or analog dials) that integrates natively with Home Assistant.

What makes this different? Unlike standalone firmwares (which are great!), this is a native ESPHome component.

  • Config in YAML: You configure it just like any other ESPHome sensor.
  • Native API: No MQTT required (unless you want it!).
  • OTA Updates: Manage it directly from your ESPHome dashboard, along with all your other ESPHome components.

:rocket: Performance & modularity

  • Optimized for ESP32: Leveraging esp-nn hardware instructions, inference time for digit recognition is now under 270ms (down from ~2.7s on older chips, but still works great on legacy ESP32 boards!).
  • Reuse the Core: The tflite_micro_helper is now a standalone component you can use to run any TFLite model in your own ESPHome projects.
  • Memory Safety: The core has been rewritten in modern C++ (C++20 compliant) with std::span and zero-copy buffers to be robust and leak-free.

:sparkles: Key Features

  • Digital & Analog (Dev State): Reads 7-segment displays AND analog dial needles (currently in development, not production ready).
  • Active Learning: Optional “Data Collector” to upload low-confidence images to help train the global model.
  • Visual Config: A web-based tool to easily draw your crop regions.

:sos: Call for Contributors (We need you!) This project has grown to over 20k lines of C++.

  • Note on Development: While the legacy code was manual, v2.0 leverages AI assistance to achieve complex dual-core optimizations and memory management. It works great, but it adds complexity!
  • We need: C++ developers to review code, power users to test edge cases, and documentation helpers.
  • Promoters: I don’t use social media (personal choice to keep my life quiet), so I rely on you to talk about this project!

If you love the idea of On-Device AI without the cloud, please check it out and consider contributing!

:link: GitHub: GitHub - nliaudat/esphome_ai_component: A component to run AI model on esphome
:open_book: Wiki: Home · nliaudat/esphome_ai_component Wiki · GitHub

2 Likes