Loading…

End-to-end deep learning-based fringe projection framework for 3D profiling of objects

Fringe projection techniques are widely used for precise three-dimensional depth profiling of objects. Existing signal processing-based fringe projection techniques measure the phase deformation of the projected fringe patterns with a sequence of operations, such as fringe denoising, fringe analysis...

Full description

Saved in:
Bibliographic Details
Published in:Computer vision and image understanding 2020-10, Vol.199, p.103023, Article 103023
Main Authors: Machineni, Rakesh Chowdary, Spoorthi, G.E., Vengala, Krishna Sumanth, Gorthi, Subrahmanyam, Gorthi, Rama Krishna Sai S.
Format: Article
Language:English
Subjects:
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Fringe projection techniques are widely used for precise three-dimensional depth profiling of objects. Existing signal processing-based fringe projection techniques measure the phase deformation of the projected fringe patterns with a sequence of operations, such as fringe denoising, fringe analysis for wrapped phase extraction, followed by phase unwrapping. However, the error induced in any of the stages leads to erroneous depth estimation. Furthermore, any aliasing in frequency domain fringe analysis and ill-posed nature of phase unwrapping limit the overall accuracy of the Fringe Projection Profilometry (FPP). To this end, unlike the traditional approaches, we propose a paradigm shift by introducing a novel end-to-end deep learning-based framework for FPP that does not need any frequency domain filtering and phase unwrapping. The proposed framework directly reconstructs the object’s depth profile from the deformed fringe itself through a multi-resolution similarity assessment convolutional neural network. We compare the performance of the proposed framework with two widely used conventional approaches. The evaluations are performed for various challenging and important scenarios such as low fringe-frequency, depth profiles with high dynamic range and noisy fringes. The results demonstrate that the proposed framework achieves promising results in all these scenarios. [Display omitted] •We propose an end-to-end deep learning-based framework for 3D profiling of objects.•This is the first end-to-end deep learning method for fringe projection profilometry.•Proposed method provides 3D profile of complex objects from single deformed fringe.•A large number of deformed fringe-absolute phase pairs are generated to train the Net.•The proposed framework outperforms existing approaches by a significant margin.
ISSN:1077-3142
1090-235X
DOI:10.1016/j.cviu.2020.103023