Novel approach for fast structured light framework using deep learning

Won-Hoe Kim

doi:10.1016/j.imavis.2024.105204

Novel approach for fast structured light framework using deep learning

2024-10-01

0

PRE

AI

DOI：10.1016/j.imavis.2024.105204

Original

Original request for help

Save

Abstract

En 中文

In structured light 3D imaging, achieving robust and accurate 3D reconstruction with a limited number of fringe patterns remains a challenge. In this study, we introduce SFNet, a symmetric fusion network that designed for high-speed, high-quality 3D surface measurement using just two fringe images. The SFNet employs separate encoders and decoders for each fringe input to estimate its phase. The two generated phase values are then utilized to reconstruct the 3D information. During the training process, we use a refined reference phase which utilizes fringe images with different frequencies. SFNet has the capability to complement the additional frequency information by fusing the feature maps extracted from each encoder. Comparative experiments and ablation studies validate the effectiveness of our proposed method. The dataset is publicly accessible on our project page https://wonhoe-kim.github.io/SFNet/.

Keywords:

Structured light

3D reconstruction

Fringe projection profilometry

Novel approach for fast structured light framework using deep learning

Abstract

Journal

Organization