Loading [a11y]/accessibility-menu.js
Self-Distilled Depth From Single-Shot Structured Light With Intensity Reconstruction | IEEE Journals & Magazine | IEEE Xplore

Self-Distilled Depth From Single-Shot Structured Light With Intensity Reconstruction


Abstract:

Depth from structured light (SL) is a mainstream approach for 3D acquisition. In this article, we propose a unique depth reconstruction method for single-shot SL systems,...Show More

Abstract:

Depth from structured light (SL) is a mainstream approach for 3D acquisition. In this article, we propose a unique depth reconstruction method for single-shot SL systems, which reconstructs the intensity image of the scene simultaneously. The merits of our method are twofold. First, the intensity image can be used to extract scene textures under ambient light from the captured image, parsing out the projected SL pattern and thus improving depth reconstruction performance. Second, the intensity information of the scene can be useful in many applications when additional RGB cameras are not available along with the SL system. The proposed method is realized by a dual-branch deep neural network for recovering depth and intensity, respectively, where the intermediate output of the intensity branch is fed into the depth branch. Specifically, we introduce a self-distillation strategy to facilitate training the network in an unsupervised manner.
Published in: IEEE Transactions on Computational Imaging ( Volume: 9)
Page(s): 678 - 691
Date of Publication: 26 June 2023

ISSN Information:

Funding Agency:


References

References is not available for this document.