Precision Comparison and Analysis of Multi‑stereo Fusion and Multi‑view Matching Based on High‑Resolution Satellite Data

High-resolution sub-meter satellite data play an increasingly crucial role in the 3D real-scene China construction initiative. Current research on 3D reconstruction using high-resolution satellite data primarily focuses on two approaches： Multi-stereo fusion and multi-view matching. While algorithms based on these two methodologies for multi-view image 3D reconstruction have reached relative maturity， no systematic comparison has been conducted specifically on satellite data to evaluate the relative merits of multi-stereo fusion versus multi-view matching methods. This paper conducts a comparative analysis of the practical accuracy of both approaches using high-resolution satellite datasets from diverse geographical regions. To ensure fairness in accuracy comparison， both methodologies employ non-local dense matching for cost optimization. Results demonstrate that the multi-stereo fusion method outperforms multi-view matching in all evaluation metrics， exhibiting approximately 1.2% higher average matching accuracy and 10.7% superior elevation precision in the experimental datasets. Therefore， for 3D modeling applications using satellite data， we recommend adopting the multi-stereo fusion approach for digital surface model （DSM） product generation.