CrackWatch — measure a wall

CrackWatch — measure a wall crack with your phone

You find a crack in a wall. Is it growing? A phone photo has no scale, so two photos months apart cannot be compared, and an expert visit costs money.

What I built • A crack gauge that needs only a bank card (ISO ID-1 size gives the scale; the card is blurred after measuring) • Width in millimetres with its uncertainty, at fixed points re-measured at every visit: "stable" or "widened by X mm" • A live camera guide in the browser that finds the card and takes the photo when it is sharp • An agent that picks the next step from the image (retake, closer photo, compare, ticket), checked by rules; a human approves every report • A precision mode with a printed calibration card, and a dated PDF report

Result • On a printed test chart photographed with a phone: mean error 0.061 mm • Bank card vs printed card on the same 13 real photos: −3.1 % to +1.3 % • Limits and failure cases published with the evaluation • Built for the OpenCV AI Competition 2026, powered by AWS

My role: Product design, computer vision, back-end and cloud, UX/UI, evaluation, demo film Stack: Python, OpenCV 5, OpenCV.js, AWS Lambda (arm64), S3, DynamoDB, Amazon Bedrock (Nova 2 Lite) Watch the demo: https://youtu.be/NJy0aBBCE90 | Code: https://github.com/blanco1er/crackwatch