Pose Estimation Using Geometric Information Based on Point Cloud of Non-cooperative Spacecraft

Spacecraft pose estimation is the foundation for space missions such as rendezvous and docking. Existing deep learning methods for pose estimation based on spacecraft point clouds lack specific designs tailored to the unique structures of spacecraft, resulting in lower estimation accuracy. This paper proposes a network that encodes geometric information from spacecraft point clouds and employs an attention mechanism to capture global structural relationships. Furthermore, a point cloud dataset is generated through simulations using spacecraft CAD models, and the proposed method is tested on this dataset. Experimental results demonstrate that the designed network achieves superior pose estimation results compared to mainstream methods.