用于城市三维重建的上海徐汇地区高精度光学卫星遥感影像数据集

A dataset of remote sensing images from high-precision optical satellites for urban three-dimensional reconstruction in Xuhui District of Shanghai

随着遥感卫星的发展和三维重建技术的进步,卫星影像在高精度城市三维建模中扮演着越来越重要的角色。神经辐射场(Neural Radiance Fields,NeRF)是一种新颖的三维重建方法,凭借高重建精度和无需依赖深度标签等特性逐渐成为三维视觉领域的研究重点。然而,目前国内外尚缺乏多视角、高分辨率、高位姿精度的卫星影像数据集,这使得利用卫星影像进行三维重建的研究受到了极大的制约。针对这一问题,本文选用2021–2023年北京三号国际合作星在上海市徐汇区上海体育馆周边地区的七景遥感数据,通过图像配准、自适应全色锐化、有理多项系数粗调整和捆集优化等处理,获得两组面积分别为0.73 km^(2)和0.25 km^(2)的多视角卫星影像数据集,并在数据来源可靠性、预处理流程规范性、分模块参数优化等多个维度对数据质量进行了严格把控。本数据集将为后续的卫星影像城市级三维重建研究——特别是NeRF系列研究提供有力的数据支持。

With the development of remote sensing satellites and the advancement of three-dimensional reconstruction technology,satellite images play an increasingly important role in high-precision city modeling.As a novel 3D reconstruction technique,Neural radiance fields(NeRF)is emerging as a research focus in the field of 3D vision,due to its high reconstruction accuracy and independence from depth labels.However,there is still a lack of multi-view,high-resolution satellite image datasets with high pose estimation accuracy domestically and abroad.This severely hinders research on 3D reconstruction using satellite imagery.To address this issue,in this study,we employed seven remote sensing scenes from the Beijing-3 satellite over the area surrounding the Shanghai Gymnasium in Xuhui District,Shanghai,from 2021-2023.Through techniques like image registration,adaptive IHS Pan-sharpening,Rational Polynomial Coefficients(RPC)coarse adjustment,and bundle adjustment optimization,we developed two datasets of multi-view satellite images covering Areas of Interest measuring 0.73 km^(2)and 0.25 km^(2).The data quality was rigorously controlled across various dimensions,including the reliability of data sources,standardization of preprocessing procedures,and optimization of parameters in modular components.This dataset is expected to provide powerful data support for subsequent research on city-scale 3D reconstruction from satellite imagery,especially those based on NeRF frameworks.