利用全局仿射模型进行卫星图像快速三维重建  

Fast 3D reconstruction of satellite images via the Global Affine Model

作　　者：陈豹 王品贺 董秋雷[1,2,3] CHEN Bao;WANG Pinhe;DONG Qiulei(National Key Laboratory of Multi-modal Artificial Intelligence Systems,Institute of Automation,Chinese Academy of Sciences,Beijing 100190,China;School of Artificial Intelligence,University of Chinese Academy of Sciences,Beijing 100049,China;Center for Excellence in Brain Science and Intelligence Technology,Chinese Academy of Sciences,Beijing 100190,China;School of Computer Science and Engineering,University of Northeastern,Shenyang 110169,China)

机构地区：[1]中国科学院自动化研究所多模态人工智能系统全国重点实验室,北京100190 [2]中国科学院大学人工智能学院,北京100049 [3]中国科学院脑科学与智能技术卓越创新中心,北京100190 [4]东北大学计算机科学与工程学院,沈阳110169

出　　处：《遥感学报》2024年第6期1576-1587,共12页NATIONAL REMOTE SENSING BULLETIN

基　　金：国家自然科学基金(编号:61991423,U1805264);中国科学院战略先导专项(编号:XDB32050100);北京市科技计划(编号:Z211100011021004)。

摘　　要：基于多视角卫星遥感图像的三维场景重建是遥感领域中一项极具挑战性的任务。针对现有方法重建速度慢的问题,本文提出了一种利用全局式仿射模型进行卫星图像快速重建方法。该方法首先将多视角卫星图像裁剪成一组局部图像块并计算相应局部场景的三维仿射点云,然后引入一种基于局部点云的全局式仿射运动矩阵估计算法计算出每个视角对应的相机仿射运动矩阵,在此基础上利用少量地面控制点恢复出场景的三维欧氏结构。在两个公共数据集上的实验结果表明,本文方法的重建速度、精度和完整性在大多数情况下均优于3种主流文献方法。Three-dimensional(3D)scene reconstruction based on multiview satellite remote sensing images is a challenging task in the field of remote sensing.Most of the existing methods either have to perform bundle adjustment repeatedly or must calculate several parameters in the rational polynomial camera model,resulting in a relatively long reconstruction time.To solve the abovementioned problems,this study considers that the local small-sized patches in large-sized satellites could be approximately modeled by the affine imaging model and proposes a fast 3D reconstruction method of satellite images based on global affine model estimation.First,the input multiview satellite images are cropped into a set of small-sized patches with overlapping regions.For each pair of patches that have a sufficient number of point correspondences from two views,the corresponding 3D affine point cloud is calculated.Second,on the basis of the obtained local point clouds,a global affine camera motion estimation algorithm is presented for calculating the affine motion matrices of the cameras corresponding to all the patches in a unified coordinate system.Finally,the obtained affine camera motion matrices and a few ground control points are utilized to recover the Euclidean scene structure.3D reconstruction is conducted for the same group of remote sensing images and all remote sensing images to verify the effectiveness of the method.The proposed solution is compared with three state-of-the-art methods(i.e.,COLMAP,S2P,and JHUAPL).Experimental results on two public datasets(i.e.,MVS3DM and DFC2019)show that the proposed method outperforms the three comparison algorithms in most cases with respect to speed,accuracy,and completeness.To verify further the reconstruction accuracy of the method,this study selects 15 complex scene areas from two public datasets,including complex scenes with built-up areas,shadow areas,and complex object areas.For 15 complex scenarios,the proposed method outperforms the three methods with respect to accuracy and comple

关 键 词：三维重建 卫星遥感图像 仿射成像模型 欧氏结构恢复 全局仿射矩阵 

分 类 号：TP701[自动化与计算机技术—检测技术与自动化装置] P2[自动化与计算机技术—控制科学与工程]