Sat-DN: Implicit Surface Reconstruction from Multi-View Satellite Images with Depth and Normal Supervision

2025-02-12

AI Summary

To address geometric distortions in implicit surface reconstruction from multi-view satellite imagery—caused by low texture, illumination variations, and façade occlusions—this paper proposes a progressive multi-resolution hash grid architecture. Methodologically, it integrates explicit depth map supervision with surface normal consistency regularization, leveraging hash-encoded neural implicit representations and a progressive frequency-uplifting training strategy to enable coarse-to-fine geometric modeling. Compared to conventional stereo matching approaches, the method achieves significantly higher geometric fidelity, especially in fine-scale details; relative to NeRF-based methods, it accelerates training substantially while mitigating overfitting. Evaluated on the DFC2019 dataset, the approach attains state-of-the-art performance, demonstrating superior building contour sharpness and planar structural accuracy over existing methods.

Technology Category

• Machine Learning: Multi-instance/Multi-view Learning
• Computer Vision: Multi-modal Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

With advancements in satellite imaging technology, acquiring high-resolution multi-view satellite imagery has become increasingly accessible, enabling rapid and location-independent ground model reconstruction. However, traditional stereo matching methods struggle to capture fine details, and while neural radiance fields (NeRFs) achieve high-quality reconstructions, their training time is prohibitively long. Moreover, challenges such as low visibility of building facades, illumination and style differences between pixels, and weakly textured regions in satellite imagery further make it hard to reconstruct reasonable terrain geometry and detailed building facades. To address these issues, we propose Sat-DN, a novel framework leveraging a progressively trained multi-resolution hash grid reconstruction architecture with explicit depth guidance and surface normal consistency constraints to enhance reconstruction quality. The multi-resolution hash grid accelerates training, while the progressive strategy incrementally increases the learning frequency, using coarse low-frequency geometry to guide the reconstruction of fine high-frequency details. The depth and normal constraints ensure a clear building outline and correct planar distribution. Extensive experiments on the DFC2019 dataset demonstrate that Sat-DN outperforms existing methods, achieving state-of-the-art results in both qualitative and quantitative evaluations. The code is available at https://github.com/costune/SatDN.

Problem

• Reconstruct terrain geometry from satellite images
• Enhance building facade details in reconstructions
• Accelerate training of high-quality satellite image models

Innovation

• Multi-resolution hash grid
• Depth and normal constraints
• Progressive training strategy