From Images2Mesh: A 3D Surface Reconstruction Pipeline for Non-Cooperative Space Objects

A team of researchers (Gopu, Quinn, Nehma, Tiwari, Ueckermann, Hinckley, McKenna) has unveiled Images2Mesh, a neural implicit surface reconstruction pipeline designed specifically for non-cooperative space objects—i.e., debris or defunct satellites that don't communicate with ground stations. Published on arXiv, the pipeline takes monocular inspection imagery (single-camera video) and outputs a 3D surface mesh, enabling operators to assess geometry and structural condition for active debris removal or on-orbit servicing missions.

The key innovation is that Images2Mesh works on real on-orbit footage, not just synthetic or hardware-in-the-loop data. Previous neural implicit methods required known camera poses and controlled lighting. The researchers demonstrated their pipeline on publicly released ISS inspection footage from the STS-119 Shuttle mission and footage of an H-IIA rocket upper stage. They found that segmentation-based background removal is absolutely essential for successful camera pose estimation—without it, direct processing on real footage fails entirely due to background variation between frames. They also incorporated photometric correction for per-frame exposure variations, noting that performance in shadowed regions varies with input illumination characteristics. The paper (25 pages, 16 figures) represents a significant step toward real-world autonomous space debris inspection.