PROHORI: Crack Width from a Phone Photo
OpenCV AI Competition 2026

Case study

AI Systems / Computer Vision

Competition entry

Dhaka, Bangladesh

AI Systems / Computer Vision · Detailed breakdown

Project overview

Why this build
mattered.

A solo entry that measures structural crack width in millimetres from one phone photo of a crack beside a ChArUco board, then tells the user what to do next.

Snapshot

OpenCV AI Competition 2026 / Competition entry / Python, OpenCV 5, Conformal prediction

PythonOpenCV 5Conformal predictionAmazon BedrockAWS Graviton4
0.070 mm MAE
widthError
0.056 mm MAE
certified
8 held-out photos
heldOut

System views

Screens and
surfaces

What shipped inside

Key features
and modules

The architecture, workflow and results that made this project useful.

Highlight

A conformal risk-control gate labels each reading certified, indicative or refused

Highlight

An agent (rules, or Amazon Bedrock Nova Lite) turns the reading into plain instructions

Highlight

Width error 0.070 mm MAE over 63 cracks on 8 held-out photos (0.056 mm on certified readings)

Delivery summary

Build notes

Project essentials

PROHORI: Crack Width from a Phone Photo

OpenCV AI Competition 2026 · Competition entry · AI Systems / Computer Vision

A conformal risk-control gate labels each reading certified, indicative or refused1
An agent (rules, or Amazon Bedrock Nova Lite) turns the reading into plain instructions2
Width error 0.070 mm MAE over 63 cracks on 8 held-out photos (0.056 mm on certified readings)3

Let's build something worth building.