A subpixel vision metrology for industrial workpiece diameter estimation

Accurate non-contact diameter measurement is essential for industrial inspection. This study introduces OneShot-SPRDN, a one-shot vision-metrology framework combining support-guided boundary localization, subpixel offset prediction and checkerboard-based calibration. The framework uses RGB, mask and masked-image inputs to reconstruct subpixel contours and estimate rod diameters. Evaluation includes an original $$0^\circ $$ reference image and four additional images of two metallic workpieces with varying diameters, illumination and camera-framing conditions. For the reference image, OneShot-SPRDN estimates a diameter of $$28.243~\textrm{mm}$$ , compared with the coordinate measuring machine (CMM) reference of $$28.237~\textrm{mm}$$ , yielding a mean absolute error (MAE) of $$6.000~\mu \textrm{m}$$ . The additional evaluations show agreement with CMM measurements across different workpiece geometries and imaging conditions, supporting the framework’s potential for industrial dimensional inspection.