project: shape from motion

Working under an Air Force SBIR phase II contract, Stellar Science sought to apply the latest shape from motion (SFM) computer vision algorithms to the space situational awareness (SSA) domain. The goal of this research project was to compute the three-dimensional (3D) shape and orientation of a satellite from two-dimensional (2D) images of it. SSA imagery presents some unique challenges and advantages for solving the shape from motion problems. By combining a number of techniques from computer vision and computer graphics, Stellar Science demonstrated the feasibility of automatically constructing a satellite model from an uncalibrated sequence of space surveillance images.

This video presents one successful test case. The computational process involves 2D feature tracking and matching, linear factorization, non-linear optimization, outlier rejection, background-foreground segmentation, volumetric shape carving, alpha shape facetization, and hardware-accelerated multi-texture mapping.