Underwater 3-D Gaussian Reconstruction Using Multiview Sonar Imaging
IEEE Sensors Letters · 2026-09 · 10 (9) · 6008604-6008604
DOI: 10.1109/lsens.2026.3716397 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Sensors Letters, 2026-09. Verfasst von Yiwen Zhou und 5 weiteren Autor:innen. Die Arbeit nennt 7 Literaturangaben.
Abstract (Originalsprache)
Underwater 3-D reconstruction using forward-looking sonars is essential for autonomous underwater vehicles, relating to their navigation, scene understanding, and operational efficiency. Underwater 3-D reconstruction can be categorized into three types: explicit volumetric techniques, generative models, and implicit neural radiance fields. However, these methods suffer from low fidelity, poor generalization, and slow training and rendering. To address these limitations, this letter proposes an explicit 3-D Gaussian reconstruction method using multiview sonar images. Our method integrates first, a feature matching method to align multiview sonar features through geometric constraints, and second, a physics-guided Gaussian splatting algorithm that optimizes Gaussian parameters using acoustic propagation characteristics. Experiments on synthetic datasets demonstrate that our approach achieves higher accuracy and faster rendering than state-of-the-art methods while preserving underwater physical consistency.
Zitieren
APA
Zhou, Y., Liu, K., Xiao, X., Zeng, X., Chen, M., & Zhang, S. (2026). Underwater 3-D Gaussian Reconstruction Using Multiview Sonar Imaging. IEEE Sensors Letters, 10(9), 6008604-6008604. https://doi.org/10.1109/lsens.2026.3716397
Vancouver
Zhou Y, Liu K, Xiao X, Zeng X, Chen M, Zhang S. Underwater 3-D Gaussian Reconstruction Using Multiview Sonar Imaging. IEEE Sensors Letters. 2026;10(9):6008604-6008604. doi:10.1109/lsens.2026.3716397
BibTeX
@article{Zhou2026,
author = {Zhou, Yiwen and Liu, Kezhong and Xiao, Xuedou and Zeng, Xuming and Chen, Mozi and Zhang, Shengkai},
title = {Underwater 3-D Gaussian Reconstruction Using Multiview Sonar Imaging},
journal = {IEEE Sensors Letters},
year = {2026},
volume = {10},
number = {9},
pages = {6008604--6008604},
doi = {10.1109/lsens.2026.3716397}
}
EUVIMED