High-precision vision-based circular holes diameter measurement method for robotic drilling

Hole accuracy significantly influences the fatigue performance of skin-panel joints, and its inspection is both fundamental and technically challenging. With advantages like efficiency, automation, and traceability, visual-based measurement method has emerged as a promising approach in aircraft manufacturing. However, high-precision vision-based measurement still faces two challenges: (1) surface stains, scratches and specular reflections violate the ideal step-edge assumption, biasing traditional statistical edge estimators toward missed detections and a low recognition rate; (2) perspective projection introduces eccentricity errors of circular targets in mainstream binocular reconstruction, because the ellipse center is traditionally taken as the actual center. To address the first challenge, a radial pixel-scanning operator samples intensity along the hole radius, applies an adaptive threshold to the edge detector, and isolates valid hole edges against the above artefacts. Robustly fitted from these edges, the ellipse is refined to sub-pixel precision by a proposed iterative algorithm that embeds a super-resolution model. To address the second challenge, an iterative reprojection-error loop first reconstructs the circle center, normal, and radius and then feeds these parameters to an embedded correction model that explicitly eliminates perspective eccentricity. Thus, a novel hole diameter measurement method with high recognition rate and accuracy is constructed. Subsequently, a binocular vision system is built and integrated into a hybrid drilling robot to carry out hole diameter measurement experiments. On an aircraft test plate, the system reaches a 100% recognition rate, a mean measuring error of 0.016 mm and a max measuring error of 0.032 mm, which confirms that the proposed method meets the ±0.05 mm accuracy requirements of aircraft manufacturing. • A vision-based hole diameter measurement method for robotic drilling is proposed. • Robust edge extraction via radial pixel scanning is developed for ellipse fitting. • A sub-pixel refinement scheme is established using image super resolution. • Hole reconstruction method embedded with eccentricity correction model is developed. • The hole recognition rate is 100%, and the maximum diameter error is 0.032 mm.