Image-based small body shape modeling using the neural implicit method

One of the objectives of cameras on spacecraft for exploration of asteroids and comets is to perform shape modeling of the small bodies. Stereo-photogrammetry (SPG) and stereo-photoclinometry (SPC) stand out as the two main image-based methods for shape modeling, used in both previous and ongoing missions. In recent years, machine learning technology has experienced rapid development and demonstrated great promise for planetary topographic modeling. However, applications to small bodies have been limited so far. In this work, we present a neural implicit shape modeling method designed specifically for small body images characterized by rapid model convergence. We select 25143 Itokawa, explored by the Hayabusa mission, as a demonstration.  The method uses a sparse set of 52 images captured by the Asteroid Multi-band Imaging Camera (AMICA). The results are consistent with models previously produced using the SPC method in terms of overall size and shape. Also, our method can effectively capture fine-scale terrain features on the surface of Itokawa. This suggests that the neural implicit method can provide a new option and insight for the 3D reconstruction of small bodies.