Spacecraft Normalized Coordinate Space for Pose and Size Estimation of Unknown Noncooperative Space Targets

Noncooperative spacecraft pose estimation is crucial for a wide range of space missions. In recent years, data-driven approaches have garnered increasing attention in this field. However, most existing methods depend on offline training with prior models of specific spacecraft, restricting their applicability to known types and failing to generalize to previously unseen targets. To overcome these limitations, we propose a Spacecraft Normalized Coordinate Space (SNCS), which serves as a unified geometric framework across various spacecraft models. For unknown targets, we learn target-specific Shape Deformation Residuals (SDR) from observed point clouds, which are used to transform the SNCS into a reference model that encodes the object's rotation information. Leveraging symmetry cues, we further reconstruct the complete target shape from single-view point clouds, enabling both pose and size estimation for unknown spacecraft. To evaluate the generalization capability of our approach, we construct and release a benchmark dataset for pose and size estimation, containing 33 diverse spacecraft types. Extensive experiments demonstrate that our method handles large shape variations effectively, delivering accurate estimation, fast response times, and strong cross-model generalization performance for unknown targets. The project page is available at: SNCS.