Abstract:Three-dimensional reconstruction utilizing few-view satellite images can effectively reduce the cost of resources. However, methods that rely on dense stereo matching suffer severe performance degradation when confronted with differences of intersection angles and illumination in such non-standard stereo images, which are limited by the matching mechanism. Neural radiance fields (NeRFs) get rid of the matching restriction by utilizing the rendering pipeline, which has made significant progress in the synthesizing novel views and 3-D reconstruction. Nevertheless, when the available images are few-view and multidate, insufficient geometric constraints and illumination variations present great challenges in generating accurate 3-D models. In this article, we proposed a 3-D reconstruction framework called few-view multidate (FVMD)-implicit surface representation (ISRe), which is based on the ISRe of NeRF, for generating complete watertight mesh and digital surface model (DSM) from FVMD satellite images. We introduce the positional encoding with adaptive frequency module to extract finer geometric information from few-view images. Then, the solar information is incorporated into the rendering process, allowing the network to distinguish color changes arising from illumination differences in multidate images. Additionally, we employ some tricks like coordinate system optimization and network restructuring for enhancing network training. Advantages of FVMD-ISRe are demonstrated through both qualitative and quantitative experiments conducted on the US3D dataset. The results highlight our framework's capacity to reconstruct accurate 3-D geometry, overcome the performance of traditional methods based on stereo matching and other NeRFs in various evaluation metrics. Code and data are available at https://github.com/HEU-super-generalized-remote-sensing/FVMD-ISRe.

3DFIRES: Few Image 3D REconstruction for Scenes with Hidden Surface

Outdoor Scene 3D Reconstruction from Multiple Point Cloud

Multi-View Depth Map Sampling for 3D Reconstruction of Natural Scene

Three-Dimensional Reconstruction of Indoor Scenes Based on Implicit Neural Representation

Visual Odometry Based 3D-Reconstruction

FVMD-ISRe: 3-D Reconstruction From Few-View Multidate Satellite Images Based on the Implicit Surface Representation of Neural Radiance Fields

I2-SDF: Intrinsic Indoor Scene Reconstruction and Editing Via Raytracing in Neural SDFs

Single-view 3D Scene Reconstruction with High-fidelity Shape and Texture

SimpleRecon: 3D Reconstruction Without 3D Convolutions

DP-MVS: Detail Preserving Multi-View Surface Reconstruction of Large-Scale Scenes

ReconFusion: 3D Reconstruction with Diffusion Priors

Data-Driven 3D Reconstruction of Dressed Humans From Sparse Views

Fine-detailed Neural Indoor Scene Reconstruction using multi-level importance sampling and multi-view consistency

Planar Surface Reconstruction from Sparse Views

Dynamic Multi-View Scene Reconstruction Using Neural Implicit Surface

Learning to Reconstruct and Understand Indoor Scenes from Sparse Views

Behind the Veil: Enhanced Indoor 3D Scene Reconstruction with Occluded Surfaces Completion

ReFu: Refine and Fuse the Unobserved View for Detail-Preserving Single-Image 3D Human Reconstruction

FastFusion: Real-Time Indoor Scene Reconstruction with Fast Sensor Motion

Res-NeuS: Deep Residuals and Neural Implicit Surface Learning for Multi-View Reconstruction

FrozenRecon: Pose-free 3D Scene Reconstruction with Frozen Depth Models