Skip to main navigation Skip to search Skip to main content

LARGE-SCALE URBAN SCENE RECONSTRUCTION USING 3D GAUSSIAN SPLATTING AND SATELLITE IMAGERY

  • Hanqing Xu
  • , Lingfei Ma*
  • , Haiyan Guan
  • , Jonathan Li
  • *Corresponding author for this work
  • East China Normal University
  • Nanjing University of Information Science & Technology
  • University of Waterloo

Research output: Contribution to journal › Conference article › peer-review

Abstract

With the rapid development of deep learning techniques, the construction of digital twin cities has become an important direction for developing digital earth. Thus, in this paper, we perform scene reconstruction using the 3D Gaussian Splatting (3DGS) method and satellite imagery in complex and large-scale urban downtown areas. Experimental results show that 3DGS significantly outperformed Neural Radiance Fields (NeRF) in both reconstruction accuracy and efficiency, improving approximately 10–13 dB in PSNR, 0.3–0.4 in SSIM, and a reduction of 0.1 in LPIPS. Moreover, 3DGS exhibited high robustness under multi-period lighting conditions, with most quantitative metric fluctuations remaining below 2%. This paper highlights the potential of 3DGS in panoramic urban modeling and multi-scene analysis for digital twins, providing efficient and reliable technical support for future urban planning and management.

Original languageEnglish
Pages (from-to)6766-6769
Number of pages4
JournalInternational Geoscience and Remote Sensing Symposium (IGARSS)
DOIs
StatePublished - 2025
Externally publishedYes
Event2025 IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2025 - Brisbane, Australia
Duration: 3 Aug 2025 → 8 Aug 2025

Keywords

  • 3D Gaussian splatting
  • digital twins
  • large-scale 3D modeling
  • neural radiance fields

Fingerprint

Dive into the research topics of 'LARGE-SCALE URBAN SCENE RECONSTRUCTION USING 3D GAUSSIAN SPLATTING AND SATELLITE IMAGERY'. Together they form a unique fingerprint.

Cite this