Rapid construction of 3-D Satellite models from limited amounts of 2-D imagery

The space domain plays a vital role in commercial, scientific and defense endeavors. Continued safe access to and use of space demands an increased awareness of the capabalities of satellites on orbit and an ability to detect changes that may indicate status and health or operator intent. The proposed research will address this challenge by studying how to rapidly and automatically extract 3D information from a set of 2D images using a new theoretical approach. Our method leverages the observation that the Fourier spectra of man-made space objects have unique patterns of spatial frequencies. These signatures are imprinted in the Fourier amplitude/phase, Knox-Thompson cross spectrum and the bispectrum by the component shapes of the satellite, e.g. solar panels, cylindrical bodies, antennae, etc. Furthermore, compared to direct imagery, the morphology of the spectrum is much less dependent on the relative arrangement of these components. Our strategy is to identify the unique features in these patterns, and analyze how changes in the feature amplitudes reveal information about the 3D geometry of the satellite. From this analysis, current techniques in machine learning, parameter estimation and 'structure from motion' will be adapted to the problem of building a 3D model from the Fourier features.