The Taub Faculty of Computer Science Events and Talks

Multiview structure recovery from a collection of images requires the recovery of the positions and orientations of the cameras relative to a global coordinate system. We present an approach to this problem that given feature correspondences across pairs of images uses the pairwise Essential Matrices to recover camera orientation by applying robust optimization using either spectral or semidefinite programming relaxations. Once the orientations are recovered we return to the point correspondences and use them directly to recover the positions of the cameras using a linear algorithm based on a novel decomposition of the Essential Matrix. Our method accounts simultaneously for all images in a collection (in contrast to sequential processing), and enables efficient and accurate computation of the camera exterior parameters. This is demonstrated on collections of images for which ground truth measurements are available, as well as Internet images.