3D Reconstruction Techniques from Satellite Imagery

Three-dimensional reconstruction from satellite imagery integrates geometric sensor modelling, dense image matching and advanced learning methods to produce digital surface models, point clouds and watertight meshes at global scale. Traditional photogrammetric pipelines rely on stereo pair matching, rational function models and bundle adjustment to derive disparity maps and elevation grids. Multi-view fusion approaches combine several depth maps with robust outlier handling and dynamic programming to manage occlusions and repetitive patterns without ground control points. Recent advances harness generative adversarial networks for image enhancement and neural implicit surface representations—using signed distance functions and volume rendering—to enable one-stage mesh generation with realistic textures. Extensions that exploit multi-date acquisitions and seasonal appearance encoding have broadened applications in urban planning, disaster response, land-cover monitoring and infrastructure management at sub-metre resolution.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Recent studies in remote sensing have demonstrated significant strides in automated mesh generation, image enhancement and temporal fusion. A neural implicit surface method employs a continuous signed distance function learned via volume rendering and a latent appearance vector to generate fully textured 3D meshes directly from multi-view imagery, showing improved accuracy and the capacity to model seasonal variations; a generative adversarial network approach refines input imagery through a perceptual loss function to enhance stereo matching performance, resulting in higher completeness and precision in point-cloud reconstructions compared to conventional pipelines; and an extended structure-from-motion pipeline for multi-date acquisitions integrates projective camera calibration, depth-map reparameterisation and skew correction to produce watertight textured meshes, outperforming state-of-the-art point-cloud methods in both completeness and median error.

3D Reconstruction Techniques from Satellite Imagery publication trend

The graph below shows the total number of articles in 3d reconstruction techniques from satellite imagery across all publications each year (not limited to Nature Index journals).

Technical terms

2.5D elevation data: Grid of surface heights without volumetric overhangs.

Multi-view stereo (MVS): Technique reconstructing 3D shape by matching features across multiple overlapping images.

Signed distance function (SDF): Continuous function encoding distance to the nearest point on a surface.

Bundle adjustment: Optimisation refining camera poses and 3D point positions by minimising reprojection errors.

Rational function model (RFM): Polynomial-ratio approximation of satellite sensor geometry.

Epipolar resampling: Rectification aligning corresponding points along epipolar lines for efficient stereo matching.

Generative adversarial network (GAN): Dual-network architecture that enhances or synthesises images through adversarial training.

References

1. Sat-Mesh: Learning Neural Implicit Surfaces for Multi-View Satellite Reconstruction. Remote Sensing (2023).
2. An Improved 3D Reconstruction Method for Satellite Images Based on Generative Adversarial Network Image Enhancement. Applied Sciences (2024).
3. 3D SURFACE RECONSTRUCTION FROM MULTI-DATE SATELLITE IMAGES. The International Archives of the Photogrammetry Remote Sensing and Spatial Information Sciences (2021).