StructGCN: A structure-aware graph convolutional network for spacecraft pose estimation

doi:10.1016/j.cja.2026.104204

Return

StructGCN: A structure-aware graph convolutional network for spacecraft pose estimation

2026-04-13

DOI：10.1016/j.cja.2026.104204

Original

Save

View PDF

Abstract

En 中文

With the growing number of spacecraft, the demand for Active Debris Removal (ADR) and On-Orbit Servicing (OOS) has significantly increased. The accurate pose estimation of the targeted spacecraft with monocular camera play a critical role in these space applications. Existing methods face an inherent trade-off among accuracy, stability, and computational efficiency, To address the challenges, we first propose a Keypoint Weight Adaptive Perception Module (KWAPM) to characterize the relative importance of each keypoint. The KWAPM effectively mitigates the influence of outliers and the occasional errors through correlation encoding and attention mechanisms. Furthermore, we propose the Keypoints Self-correction Network (KSN), based on the incorporation of the spacecraft’s prior 3D model. It implicitly models the global spatial relationships among keypoints to improve localization accuracy. Finally, a novel Dual-Encoder Cross Attention Graph Convolutional Network (DECA-GCN) is introduced. By encoding semantic keypoints as graph nodes and aggregating the neighborhood feature between 2D keypoint and 3D priorities, we enable efficient and high-accuracy spacecraft pose estimation. Extensive experiments on the Spacecraft Pose Estimation Dataset (SPEED), SPEED+, and SWISSCUBE datasets demonstrate that the proposed method achieves competitive performance compared to state-of-the-art approaches, outperforming two-stage methods while maintaining faster inference speed than existing single-stage algorithms. Experimental results on the embedded device demonstrate that the proposed method achieves twice the inference speed of state-of-the-artapproaches while maintaining high estimation accuracy.

Keywords:

End-to-End

Graph neural networks

Monocular Vision

On-Orbit Servicing

Spacecraft pose estimation

AI Summary

Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.