JLSDNet: Jointly Learning Structured Light and Depth Estimation for Active Stereo
IEEE Transactions on Circuits and Systems for Video Technology · 2026-11 · 36 (11) · 16454-16468
DOI: 10.1109/tcsvt.2026.3718643 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Circuits and Systems for Video Technology, 2026-11. Verfasst von Xiao Yang und 6 weiteren Autor:innen. Die Arbeit nennt 70 Literaturangaben. Gefördert von: National Key Research and Development Project of China, National Natural Science Foundation of China, Major Program of National Natural Science Foundation of China.
Abstract (Originalsprache)
Active stereo systems based on structured light have emerged as a promising solution for high-accuracy 3D perception. By projecting specially designed patterns onto object surfaces, these systems provide reliable correspondence cues that facilitate dense depth reconstruction. However, existing approaches typically employ pre-defined projection patterns that are decoupled from the scene content, device characteristics, and depth estimation algorithms, which severely limits the overall performance of active stereo systems. In this work, we present a novel end-to-end framework, termed JLSDNet, that jointly learns structured light patterns and depth estimation for active stereo. Specifically, 1) we propose an Adaptive Structured-light Generation Network (ASGNet) that produces optimized projection patterns tailored to each scene. 2) We introduce a hybrid differentiable imaging model that integrates hardware in the forward pass and an optical simulation model in the backward pass, enabling efficient and accurate gradient propagation through active stereo systems. 3) We design a novel Disparity-Attention Depth Estimation Network (DANet) that fuses structured light and RGB information to build a more efficient cost volume, which suppresses redundant information and enhances depth estimation accuracy. Experiments demonstrate that our method outperforms state-of-the-art methods in challenging scenes.
Förderung
- National Key Research and Development Project of China
- National Natural Science Foundation of China
- Major Program of National Natural Science Foundation of China
- 111 Project
Zitieren
APA
Yang, X., Jia, T., Liu, Y., Li, X., Chen, D., Deng, S., & Wang, H. (2026). JLSDNet: Jointly Learning Structured Light and Depth Estimation for Active Stereo. IEEE Transactions on Circuits and Systems for Video Technology, 36(11), 16454-16468. https://doi.org/10.1109/tcsvt.2026.3718643
Vancouver
Yang X, Jia T, Liu Y, Li X, Chen D, Deng S, et al. JLSDNet: Jointly Learning Structured Light and Depth Estimation for Active Stereo. IEEE Transactions on Circuits and Systems for Video Technology. 2026;36(11):16454-16468. doi:10.1109/tcsvt.2026.3718643
BibTeX
@article{Yang2026,
author = {Yang, Xiao and Jia, Tong and Liu, Yizhe and Li, Xiaofang and Chen, Dongyue and Deng, Shizhuo and Wang, Hao},
title = {JLSDNet: Jointly Learning Structured Light and Depth Estimation for Active Stereo},
journal = {IEEE Transactions on Circuits and Systems for Video Technology},
year = {2026},
volume = {36},
number = {11},
pages = {16454--16468},
doi = {10.1109/tcsvt.2026.3718643}
}
EUVIMED