Hierarchical Warp Stereo

Abstract

This paper describes a new technique for use in the automatic production of digital terrain models from stereo pairs of aerial images. This technique employs a coarse-to-fine hierarchical control structure both for global constraint propagation and for efficiency. By the use of disparity estimates from coarser levels of the hierarchy, one of the images is geometrically warped to improve the performance of the cross-correlation-based matching operator. A newly developed surface interpolation algorithm is used to fill holes wherever the matching operator fails. Experimental results for the Phoenix Mountain Park data set are presented and compared with those obtained by ETL.