3D structure recovery from a collection of 2D photos is a classical problem in computer vision that requires the estimation of the camera orientations and locations, i.e. the camera motion. For a large, irregular collection of images, high quality camera location estimation turns out to be a complex, ill-conditioned problem. In our work, we introduce a computationally efficient algorithm based on robust-distributed convex programming for location estimation. We introduce the concept of "global parallel-rigidity" to the location estimation problem, show how to extract maximally global parallel-rigid components of the available location information and formulate a stable semidefinite program (SDP) for high levels of pairwise direction information noise. For large sets of images, we also formulate fast convex programs to produce the global camera location solution from partial solutions. This is a joint work with Amit Singer (Princeton University) and Ronen Basri (Weizmann Institute of Science).
Camera Motion Estimation by Convex Programming
May 2 2013 - 3:45pm
102A McDonnell Hall