Defect inspection of highly reflective objects using adaptive-projection structured light and PointNet++
Optics Express · 24.08.2026 · 34 (17) · 30812
DOI: 10.1364/oe.601355 ↗Kurz gefasst
Fachartikel in der Zeitschrift Optics Express, online erschienen am 10.08.2026. Verfasst von Jingsong Zeng und 2 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 23 Literaturangaben. Gefördert von: Scientific Research Project of Department of Education of Jilin Province.
Abstract (Originalsprache)
Highly reflective surfaces often cause fringe saturation in structured-light measurement, which reduces reconstruction completeness and degrades subsequent defect segmentation. To address this issue, this paper presents an integrated framework that combines adaptive structured-light projection with an improved PointNet++ network. A high-precision fringe projection system was built with average reprojection errors of 0.04 pixels for the camera and 0.05 pixels for the projector, demonstrating good calibration accuracy. For reflective regions, the optimal pixel-wise projection intensity is estimated through multi-level grayscale probing and local response fitting, enabling adaptive fringe generation that suppresses overexposure while preserving usable modulation. For point-cloud defect segmentation, PointNet++ is enhanced by strided KNN grouping, multi-branch geometric feature fusion, and lightweight Transformer self-attention. Experiments show that the proposed network achieves 81.18% mIoU and 90.76% OA on Real3D-AD, and 82.67% mIoU and 98.92% OA on the self-built reflective-surface defect dataset. These results demonstrate the effectiveness of the proposed method for reflective-surface reconstruction and defect segmentation.s
Förderung
- Scientific Research Project of Department of Education of Jilin Province
Zitieren
APA
Zeng, J., Meng, Y., & Huang, J. (2026). Defect inspection of highly reflective objects using adaptive-projection structured light and PointNet++. Optics Express, 34(17), 30812. https://doi.org/10.1364/oe.601355
Vancouver
Zeng J, Meng Y, Huang J. Defect inspection of highly reflective objects using adaptive-projection structured light and PointNet++. Optics Express. 2026;34(17):30812. doi:10.1364/oe.601355
BibTeX
@article{Zeng2026,
author = {Zeng, Jingsong and Meng, Yanli and Huang, Jipeng},
title = {Defect inspection of highly reflective objects using adaptive-projection structured light and PointNet++},
journal = {Optics Express},
year = {2026},
volume = {34},
number = {17},
pages = {30812},
doi = {10.1364/oe.601355}
}
EUVIMED