Multi-View Point Cloud Registration Guided by 3D Bounding Boxes for Collaborative Reconstruction of Non-Cooperative Space Targets

To address the challenges of collaborative reconstruction for non-cooperative space targets, this paper proposes a multi-view point cloud registration method guided by 3D bounding boxes. The proposed approach effectively mitigates issues such as mismatched point pairs and accumulated errors, which are commonly encountered during the multi-view registration of large-scale symmetric targets. First, a bounding box detection network is employed to identify and recover the relative positions and depths of target key points, providing prior pose information for unordered viewpoints and enabling rapid coarse registration. Next, an improved weighted optimization algorithm for matched point pairs is introduced to refine the registration process. Finally, a global optimization strategy is applied to reduce accumulated errors during multi-view registration, significantly improving overall registration accuracy. Experimental results demonstrate that the proposed method can accurately reconstruct target point cloud models, achieving superior accuracy and robustness compared to existing methods. This approach holds significant potential for applications such as space docking and target capture.