文献类型：会议论文

英文题名：Sparse Reconstruction of Uncooperative Space Targets Based on LightGlue

作者：Zhao, Yibin[1];Yi, Jianjun[1];Pan, Yihan[1];Chen, Liwei[1]

机构：[1]East China Univ Sci & Technol, Shanghai 200237, Peoples R China

会议论文集：17th International Conference on Machine Learning and Computing-ICMLC-Annual

会议日期：FEB 14-17, 2025

会议地点：Guangzhou, PEOPLES R CHINA

语种：英文

外文关键词：uncooperative space target; 3D reconstruction; feature; matching; SfM; Transformers mlp

摘要：The 3D reconstruction of uncooperative space targets plays many important roles in space operations. Based on the problems of strong interference and large scale, we proposed a sparse 3D reconstruction method for multi-view 2D images. We built a minimum SFM (Structure-from-Motion) system as a comparison benchmark. First, feature extraction and matching are performed on all images, and sequential images are obtained based on consensus features. Then incremental sparse reconstruction is performed, and finally point cloud postprocessing filtering is performed. In order to improve the density and accuracy of reconstruction points, LightGlue neural network is used for 2D feature matching. It has been verified that compared with the existing brute force matching, this method is more suitable for uncooperative space targets, and has obvious advantages in reconstruction accuracy and density.