Construction and refinement of panoramic mosaics with global and local alignment

Sixth International Conference on Computer Vision (ICCV'98) |

Published by IEEE Computer Society

This paper presents a new approach to computing depth maps from a large collection of images where the camera motion has been constrained to planar concentric circles. We resample the resulting collection of regular perspective images into a set of multiperspective panoramas, and then compute depth maps directly from these resampled images. Only a small number of multiperspective panoramas is needed to obtain a dense and accurate 3D reconstruction, since our panoramas sample uniformly in three dimensions: rotation angle, inverse radial distance, and vertical elevation. Using multiperspective panoramas avoids the limited overlap between the original input images that causes problems in conventional multi-baseline stereo. Our approach differs from stereo matching of panoramic images taken from different locations, where the epipolar constraints are sine curves. For our multiperspective panoramas, the epipolar geometry, to first order, consists of horizontal lines. Therefore, any traditional stereo algorithm can be applied to multiperspective panoramas without modification. Experimental results show that our approach generates good depth maps that can be used for image-based rendering tasks such as view interpolation and extrapolation.