Algorithm Modules
Six production-ready vision algorithms plus a Plugin SDK — all accessible from C/C++ with no OpenCV or PyTorch dependency in your host application.
Geometric Fitting
Circle Fitting
Geometric FittingAnnular ROI + 24-sector centroid sampling + RANSAC + Levenberg–Marquardt refinement. Centre accuracy < 0.1 px. Supports both Canny and Devernay sub-pixel edge modes.
Line Fitting
Geometric FittingRotated-rectangle ROI with segmented centroid sampling along the line direction. RANSAC + least-squares refinement. Endpoint accuracy < 0.2 px.
Ellipse Fitting
Geometric FittingElliptical annular ROI with eccentricity constraint filtering. Robust to partial occlusion and non-uniform illumination.
Rectangle Fitting
Geometric FittingFour-edge detection via Hough / RANSAC with right-angle orthogonality constraint. Outputs four corner points in sub-pixel coordinates.
Feature Matching
Shape Template Matching
Feature MatchingGradient-orientation based rotation-invariant template matching. Robust to illumination changes and partial occlusion. No retraining needed.
Defect Detection
PatchCore Defect Detection
Defect DetectionDINOv2 + PatchCore unsupervised anomaly detection. Train on 5–10 normal images only — zero defect samples required. Outputs per-pixel heatmap. C++ inference via LibTorch.
Integration
Plugin SDK
IntegrationPure-C API for integrating VisionLab as a subprocess plugin. Windows shared-memory IPC with < 2 ms round-trip. Supports sync, async batch, and embedded-UI modes. No host-side OpenCV or PyTorch dependency.