Deep learning-based fringe projection profilometry driven by a hybrid dataset
Applied Optics · 01.08.2026 · 65 (22) · 7481
DOI: 10.1364/ao.599104 ↗Kurz gefasst
Fachartikel in der Zeitschrift Applied Optics, online erschienen am 27.07.2026. Verfasst von Jianhua Wang und 4 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 43 Literaturangaben. Gefördert von: Natural Science Foundation of Shandong Province.
Abstract (Originalsprache)
With the development of artificial intelligence, the integration of deep learning (DL) and fringe projection profilometry (FPP) is leading the revolution of optical three-dimensional (3D) reconstruction technology, making it possible to achieve high-precision phase demodulation from a single-frame fringe. To improve the generalization of the network, the fringe projection 3D reconstruction method based on DL requires large-scale and high-quality data samples. To obtain massive samples at low cost, we establish a digital twin of FPP to generate simulation samples. Subsequently, we combine simulation samples with a small number of real captured samples to construct a hybrid dataset, thereby improving the phase demodulation accuracy and generalization ability of the network. The experimental results show that compared with methods that only train simulation or real samples, the proposed method effectively improves phase accuracy and robustness.
Förderung
- Natural Science Foundation of Shandong Province
Zitieren
APA
Wang, J., Ji, Y., Ma, Q., Sun, H., & Shao, T. (2026). Deep learning-based fringe projection profilometry driven by a hybrid dataset. Applied Optics, 65(22), 7481. https://doi.org/10.1364/ao.599104
Vancouver
Wang J, Ji Y, Ma Q, Sun H, Shao T. Deep learning-based fringe projection profilometry driven by a hybrid dataset. Applied Optics. 2026;65(22):7481. doi:10.1364/ao.599104
BibTeX
@article{Wang2026,
author = {Wang, Jianhua and Ji, Yixin and Ma, Qinkui and Sun, Huixin and Shao, Tong},
title = {Deep learning-based fringe projection profilometry driven by a hybrid dataset},
journal = {Applied Optics},
year = {2026},
volume = {65},
number = {22},
pages = {7481},
doi = {10.1364/ao.599104}
}
EUVIMED