Duration: 12 months
For in-orbit rendezvous applications, inspection, or debris capture, determining the 3D geometry of the target is essential. Pre-existing 3D models may be incomplete or unavailable, especially for unknown targets. A method for characterizing the 3D geometry of a target from camera images captured under realistic operational conditions would be highly beneficial.
Current and planned imaging systems often produce noisy images of small targets at low temporal frequencies. To address these challenges, our project aims to develop a method for computing a coarse 3D mesh model of a target from single-shot images using deep learning techniques and an original approach.
The neural network model will be trained on an extensive dataset of randomized satellite images, including procedurally generated models. The solution viability will be readily tested for consideration on future servicing and SSA (Space Situational Awareness) missions.
