How it works
- Quality gatesharpness and exposure of the frame; a blurred frame ends in a retake request.
- Scale and planeOpenCV 5
aruco.ArucoDetector finds the marker with sub-pixel corners; a homography makes the surface fronto-parallel at the marker's native pixel density. The marker edge doubles as a blur probe (10–90 % edge rise gives the optical sigma).
- Finda small U-Net (1.56 M parameters, ONNX) runs in the OpenCV 5 DNN engine on a 1024 px view; a classical multi-scale Hessian ridge detector runs as an independent second opinion.
- Measureon the full-resolution rectified image: centreline by thinning, grey-level profiles across the crack every few pixels with
cv.remap, full width at half depth, then a blur model inverts the optics. Widths the optics cannot resolve are reported as upper bounds.
- Decidethe widest confirmed crack (95th percentile along its length) against the ACI 224R-01 guideline for the chosen exposure, with uncertainty. Too close to call, unconfirmed, or unresolvable cases go to a person.
ACI 224R-01 Table 4.1 is a design guideline for new construction under service loads. ACI notes that a portion of cracks will exceed it. CrackRuler uses it to route a finding, never to certify a structure as safe. Photos you upload are processed in memory and are not stored; only the report and the annotated image are kept for 7 days.