| Abstract [eng] |
Optical sensor data serves as the primary information source during the early approach phases of space missions, preceding the operational range of active instruments like laser altimeters. While photogrammetry offers a pathway for early shape and volume modelling, its application in space is constrained by radiometric and kinematic conditions. This research investigates the reliability and accuracy of Structure-from-Motion and Multi-View Stereo algorithms under these specialised constraints. The study utilises a three-phase progressive methodology. First, the computational pipeline was validated using procedurally generated synthetic asteroid model. Second, realistic space conditions were simulated, incorporating single-source illumination, high-contrast shadows, and restricted fly-by trajectories to stress-test the algorithms. Finally, a heuristic filtering and processing pipeline was applied to real optical imagery of asteroid (101955) Bennu, extracted from the OSIRIS-REx mission archive. Reconstructed models were quantitatively evaluated against high-precision laser altimeter data using the Cloud-to-Mesh distance metric. Synthetic simulations demonstrated a fundamental algorithmic trade-off regarding input data density. While maximising image resolution universally improves geometric accuracy, increasing the number of captured frames presents a distinct duality. Higher frame density significantly enhances overall surface coverage and the fidelity of the global base geometry, yet it simultaneously elevates the average statistical deviation by introducing localised structural noise in visually ambiguous regions. When applied to archival data, orbital phase sequence failed to reconstruct due to dynamic shadows and insufficient baseline frames overlap. However, the approach phase sequence, where spacecraft's linear motion combined with the asteroid's rotation to create virtual orbit effect, yielded a successful model. The final photogrammetric model achieved a mean geometric deviation of around 2 meters compared to the laser altimeter reference, representing approximately 0.4 % of the asteroid's average diameter. The findings confirm that while monocular photogrammetry lacks the high-frequency topographic precision required for landing site selection, it successfully captures global shape and relative volume, making it a viable tool for early-phase mission planning and initial object characterisation. |