Omnidirectional 3D reconstruction using stereo multi-perspective panoramas

In this paper, we propose a novel method for omnidirectional 3-D reconstruction using multi-perspective panoramas. We use two large collections of images taken by parallel stereo cameras whose motions are constrained to a planar concentric circle. The two collections of regular perspective images are resampled into four multi-perspective panoramas. Then we compute a depth map from three pairs of the panoramas using multi-baseline algorithm with three types of epipolar constraints, that is horizontal, vertical and combination of them. The validity of the proposed method is shown by experimental results using both synthesized and real images.