Efficient Multi-View Stereo for Space Target with Mamba-Based Cost Aggregation

Space on-orbit service technologies have witnessed rapid development in recent years, with 3D reconstruction technology for non-cooperative space target plays a pivotal role in the autonomous execution of complex space operation. Given that existing pose measurement methods mostly rely on known 3D models, a faster approach to achieve high-precision 3D reconstruction from a monocular image sequence is crucial for achieving high-speed and high-accuracy pose measurement of unknown space objects that may undergo deformation. Benefiting from the linear complexity and rapid inference advantages of the Mamba architecture, we propose MambaMVS, a novel end-to-end Multi-View Stereo (MVS) network using Mamba for efficient cost aggregation. In particular, we design a Mamba-based cost aggregation network named MCV Module. Experiments demonstrate that our method achieves state-of-the-art performance on the Customized Satellite Dataset while exhibiting a superior speed-accuracy trade-off on the public DTU dataset, validating the effectiveness of the proposed method, which offers new technical possibilities for building real-time perception layers in autonomous space operation systems.