Three-Dimensional Reconstruction of a Resident Space Object through Structure From Motion

Abstract

This project aims to build a system that can produce a three-dimensional (3D) model of a Resident Space Object (RSO) based on images taken by a satellite flying in formation with the RSO. The 3D reconstruction is built using Structure from Motion, which projects multiple points from two-dimensional (2D) images into 3D space and minimizes the reprojection error through camera resectioning and optimization algorithms. Contrary to most existing reconstruction processes, this project does not make use of depth-finding through laser scanning to locate the 3D position of points. Instead, it relies on imaging the RSO through camera resectioning. We conducted an analysis of the accuracy and robustness of our 3D reconstruction system and com-pared this directly to published results of laser scan reconstruction systems. We found that while our 3D reconstruction system is capable of reconstructing distinct RSO models with a 5% error and in a smaller time frame than laser scan systems, it is more sensitive to environmental noise and uses larger computing power than initially expected. Thus, despite a greater cost-efficiency associated with reduced hardware, 3D reconstruction through camera imaging is not yet sufficiently developed to produce precise and robust results and outperform laser scan systems.