The quick and accurate retrieval of an object's depth from a single-shot fringe pattern in fringe projection profilometry has been a topic of ongoing research. In recent years, with the development of deep learning, a deep learning technique to FPP for single-shot 3D measurement is being used. To improve the accuracy of depth estimation from a single-shot fringe pattern, we propose the depthwise separable Dilation Inceptionv2-UNet (DD-Inceptionv2-UNet) by adjusting the depth and width of the network model simultaneously. And we evaluate the model on both simulated and experimental datasets. The experimental results show that the error between the depth map predicted by the proposed method and the label is smaller, and the depth curve map is closer to the ground truth. And on the simulated dataset, the MAE of the proposed method decreased by 35.22%, compared to UNet. On the experimental dataset, the MAE of the proposed method decreased by 34.62%, compared to UNet. The proposed method is relatively outstanding in both quantitative and qualitative evaluations, effectively improving the accuracy of 3D measurement results from a single-shot fringe pattern.
Download full-text PDF |
Source |
|---|---|
| http://dx.doi.org/10.1364/AO.504023 | DOI Listing |
Publication Analysis
Top Keywords
Similar Publications
This Letter introduces a novel, to the best of our knowledge, calibration method for structured light systems that simplifies the calibration process and reduces time consumption. We combine vertical and horizontal fringe patterns into a single composite pattern and retrieve the bidirectional phase based on Fourier transform profilometry (FTP). The entire calibration process becomes faster and more simplified by capturing only a single-shot pattern.
View Article and Find Full Text PDFDeep Learning for Single-Shot Structured Light Profilometry: A Comprehensive Dataset and Performance Analysis.
J Imaging
July 2024
Industrial Vision Lab (InViLab), Faculty of Applied Engineering, Campus Groenenborger, University of Antwerp, Groenenborgerlaan 179, 2020 Antwerp, Belgium.
Article Synopsis
- Single-shot deep learning-based structured light profilometry (SS-DL-SLP) offers fast and reliable 3D measurement, but creating large training datasets for it is a challenge due to practical constraints.
- The paper introduces a dataset with over 10,000 examples created by 3D-printing a calibration target and recording both height profiles and fringe patterns.
- The researchers analyzed various neural networks for accuracy and robustness, and they made their dataset and code publicly available to encourage further research and development in DL-based techniques for SS-DL-SLP.
Refractive index measurements are critical for characterizing the properties of hypersonic flows, but moderate- to high-pressure experiments require alternative methods to traditional interferometric fringe counting. In this work, we introduce a novel, to the best of our knowledge, multi-wavelength phase-correlation interferometric technique to estimate the refractive index changes across nearly discrete shock wave boundaries and also simultaneously capture optical dispersion and vibrational relaxation times. By comparing the interference pattern of three or more wavelengths against each other, the refractive index can be accurately determined.
View Article and Find Full Text PDFThe accuracy of phase demodulation has significant impact on the accuracy of fringe projection 3D measurement. Currently, researches based on deep learning methods for extracting wrapped phase mostly use U-Net as the subject of network. The connection method between its hierarchies has certain shortcomings in global information transmission, which hinders the improvement of wrapped phase prediction accuracy.
View Article and Find Full Text PDFSingle-Shot 3D Reconstruction via Nonlinear Fringe Transformation: Supervised and Unsupervised Learning Approaches.
Sensors (Basel)
May 2024
Department of Mechanical Engineering, School of Engineering, The Catholic University of America, Washington, DC 20064, USA.
The field of computer vision has been focusing on achieving accurate three-dimensional (3D) object representations from a single two-dimensional (2D) image through deep artificial neural networks. Recent advancements in 3D shape reconstruction techniques that combine structured light and deep learning show promise in acquiring high-quality geometric information about object surfaces. This paper introduces a new single-shot 3D shape reconstruction method that uses a nonlinear fringe transformation approach through both supervised and unsupervised learning networks.
View Article and Find Full Text PDFWant AI Summaries of new PubMed Abstracts delivered to your In-box?
Enter search terms and have AI summaries delivered each week - change queries or unsubscribe any time!