Gaussian Splatting Could Be the Next Big Innovation in Satellite Photogrammetry

Faster 3D Reconstruction for Earth Observation Applications

Photogrammetry has long relied on stereo imagery, bundle adjustment, and dense matching techniques to reconstruct three-dimensional surfaces. More recently, Neural Radiance Fields (NeRFs) introduced powerful new capabilities but often came with significant computational costs.

A 2025 CVPR paper titled “Gaussian Splatting for Efficient Satellite Image Photogrammetry” proposes a compelling alternative that may dramatically reduce processing times while maintaining high reconstruction accuracy.

Source: CVPR 2025 Paper

What Is Gaussian Splatting?

Gaussian Splatting is a relatively new 3D scene representation technique that models environments using collections of Gaussian primitives instead of conventional meshes or volumetric neural fields.

The approach offers:

• Rapid training.
• Efficient rendering.
• High-quality visualisation.
• Reduced computational requirements.

Originally developed for computer vision applications, researchers are now adapting the technique to remote sensing workflows.

Why Satellite Photogrammetry Needs Innovation

Satellite image archives continue to expand rapidly, generating enormous volumes of data.

Traditional approaches often face challenges including:

• Long processing times.
• High computational costs.
• Difficulties handling temporal datasets.
• Limited scalability.

The new Earth Observation Gaussian Splatting framework adapts the technique specifically for satellite imagery by incorporating:

• Radiometric correction.
• Shadow modelling.
• View consistency constraints.
• Opacity regularisation.

Key Findings

According to the authors, the technique achieves state-of-the-art performance for certain satellite reconstruction tasks while requiring only a fraction of the computational effort associated with NeRF-based approaches.

This could significantly reduce barriers to:

• Large-area terrain modelling.
• Urban reconstruction.
• Change detection.
• Digital Surface Model (DSM) generation.

Implications for Geospatial Professionals

For surveyors, photogrammetrists, and geospatial data providers, processing efficiency translates directly into operational benefits.

Faster reconstruction pipelines can enable:

• Reduced cloud computing costs.
• More frequent updates.
• Near real-time processing workflows.
• Improved scalability for national mapping programs.

While traditional photogrammetry remains essential for survey-grade deliverables, Gaussian Splatting may emerge as a valuable addition to the industry toolkit.

The coming years will reveal how quickly these techniques move from research papers into commercial reality.