Reproduce “Joint attitude estimation and 3D neural reconstruction of non-cooperative space objects” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Clément Forray, Pauline Delporte, Nicolas Delaygue, Florence Genin, Dawa Derksen
A joint NeRF-based method reconstructs a non-cooperative spacecraft while estimating its unknown attitude from monochromatic image sequences. In a controlled simulation of a Galileo satellite completing slightly more than one rotation, the global uniform-rotation model reaches a median pose error of 1.6° and a maximum error of 3°, while producing comparable novel-view quality to known-pose and independently optimized baselines.
Key insights
- Constraining all views with a physically motivated uniform-rotation model reduces pose ambiguity and is more stable than optimizing one unconstrained rotation per image.
- Joint pose and geometry estimation preserves image-level reconstruction quality: the global model achieves 33.12 dB PSNR and 0.0158 LPIPS, compared with 34.27 dB and 0.0123 LPIPS for the known-pose baseline.
- Feature matching alone is poorly suited to the simulated imagery, with SfM reaching an 83° median attitude error, whereas the global optimization reaches 1.6° without a pose initialization prior.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.