Reading Your Gas Meter Is Dumb. So I Automated It

In this video, I take on one of my most ambitious smart-home projects yet: teaching a Raspberry Pi to read my gas meter using machine learning. Every month I had to crawl into a cupboard under my stairs, point a torch at the meter, and read off eight tiny digits for my energy supplier. So instead, I put a camera in the cupboard and trained a neural network to do it for me — building everything from first principles, including how a computer actually "sees" an image, how a YOLO object detector finds and classifies each digit in a single pass, and how transfer learning let me reach 99.5% accuracy from just a few thousand of my own photos. The end result is a fully local, hands-off gas meter that reads itself, sanity-checks itself with a couple of physics-based guardrails, posts straight to Home Assistant, and submits the reading to the gas company on its own — no crawling, no torch, no manual readings ever again. --- 👋 Follow me → https://linktr.ee/keenontech 📦 Patreon →   / keenontech   ☕ Kofi → https://ko-fi.com/cian911 🙇 Github Sponsors → https://github.com/sponsors/Cian911/d... --- ⚡️ Project links: Github Project: https://github.com/Cian911/smart-gas-... Ultralytics YOLOv8: https://github.com/ultralytics/ultral... Raspberry Pi 4: https://amzn.to/4gwipmb Raspberry Pi Camera Module 3: https://amzn.to/4oS6d0S Raspberry Pi 4B Case & Camera Mount: https://bit.ly/4xXzhbt --- Content: 00:00 - Intro 00:45 - The Build 02:06 - How a computer actually sees an image 04:07 - How the neural network works 06:17 - Transfer learning & training 09:07 - Deployment (Pi + Home Assistant) 11:02 - Learnings 11:52 - Outro