Enhanced Cross-Modal Point Cloud Completion Framework

Point cloud completion is crucial for 3D reconstruction and spatial perception, yet in complex scenes like Earth-Moon space, occlusions, illumination variations, and viewpoint constraints lead to severe data loss, hindering high-quality reconstruction. Traditional point cloud completion methods rely on single-modal point clouds or shallow image-point cloud fusion, struggling to capture complex geometric structures and limiting reconstruction accuracy. To address these limitations, a multi-view image-assisted cross-modal point cloud completion method is proposed, centered on the Enhanced Cross-Modal Optimization Module (ECMO). ECMO leverages the attention mechanisms of the Cross-Modal Fusion Transformer (CMFT) to achieve deep alignment of visual and geometric features, while the PointTransformer optimizes global shape consistency, generating detail-rich point clouds. Evaluations on the ShapeNetViPC dataset demonstrate that this method significantly reduces Chamfer Distance, outperforming the baseline ViPC model, providing an efficient solution for 3D modeling.