Learning Dense Stereo Matching for Digital Surface Models from Satellite Imagery

TL;DR

This paper introduces a neural network tailored for generating Digital Surface Models from satellite imagery, addressing the unique challenges of satellite stereo reconstruction and demonstrating improved results over existing methods.

Contribution

The paper presents a novel neural network architecture and training scheme specifically designed for satellite imagery DSM generation, a largely unexplored area in deep learning.

Findings

Models produce smooth and boundary-preserving DSMs

The approach outperforms existing methods in satellite stereo matching

First deep learning application for satellite DSM generation

Abstract

Digital Surface Model generation from satellite imagery is a difficult task that has been largely overlooked by the deep learning community. Stereo reconstruction techniques developed for terrestrial systems including self driving cars do not translate well to satellite imagery where image pairs vary considerably. In this work we present neural network tailored for Digital Surface Model generation, a ground truthing and training scheme which maximizes available hardware, and we present a comparison to existing methods. The resulting models are smooth, preserve boundaries, and enable further processing. This represents one of the first attempts at leveraging deep learning in this domain.