Version 1
: Received: 23 September 2024 / Approved: 24 September 2024 / Online: 25 September 2024 (11:32:16 CEST)
How to cite:
Ngcobo, K.; Bhengu, S.; Mudau, A.; Thango (Y2-rated Researcher), B.; Matshaka, L. From Single Shot to Structure: End-to-End Network based Deflectometry for Specular Free-Form Surface Reconstruction. Preprints2024, 2024091913. https://doi.org/10.20944/preprints202409.1913.v1
Ngcobo, K.; Bhengu, S.; Mudau, A.; Thango (Y2-rated Researcher), B.; Matshaka, L. From Single Shot to Structure: End-to-End Network based Deflectometry for Specular Free-Form Surface Reconstruction. Preprints 2024, 2024091913. https://doi.org/10.20944/preprints202409.1913.v1
Ngcobo, K.; Bhengu, S.; Mudau, A.; Thango (Y2-rated Researcher), B.; Matshaka, L. From Single Shot to Structure: End-to-End Network based Deflectometry for Specular Free-Form Surface Reconstruction. Preprints2024, 2024091913. https://doi.org/10.20944/preprints202409.1913.v1
APA Style
Ngcobo, K., Bhengu, S., Mudau, A., Thango (Y2-rated Researcher), B., & Matshaka, L. (2024). From Single Shot to Structure: End-to-End Network based Deflectometry for Specular Free-Form Surface Reconstruction. Preprints. https://doi.org/10.20944/preprints202409.1913.v1
Chicago/Turabian Style
Ngcobo, K., Bonginkosi Thango (Y2-rated Researcher) and Lerato Matshaka. 2024 "From Single Shot to Structure: End-to-End Network based Deflectometry for Specular Free-Form Surface Reconstruction" Preprints. https://doi.org/10.20944/preprints202409.1913.v1
Abstract
Deflectometry is a key component in the precise measurement of specular (mirrored) surfaces, however traditional methods often lack an end-to-end approach that performs 3D reconstruction in a single shot with high accuracy and generalizes across different free-form surfaces. This paper introduces a novel deep neural network (DNN)-based approach for end-to-end 3D reconstruction of free-form specular surfaces using single-shot deflectometry. Our proposed network, VUDNet, innovatively combines discriminative and generative components to accurately interpret orthogonal fringe patterns and generate high-fidelity 3D surface reconstructions. By leveraging a hybrid architecture integrating a Variational Autoencoder (VAE) and a modified U-Net, VUDNet excels in both depth estimation and detail refinement, achieving superior performance in challenging environments. Extensive data simulation using Blender, leading to a dataset which we are making available, ensures robust training and enables the network to generalize across diverse scenarios. Experimental results demonstrate the strong performance of VUDNet, setting a new standard for 3D surface reconstruction. The implementation of VUDNet is available on GitHub at https://github.com/hadious/VUDNet.
Keywords
deflectometry; specular surface reconstruction; deep neural networks; single-shot measurement; data simulation; variational autoencoder (VAE)
Subject
Business, Economics and Management, Business and Management
Copyright:
This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.