Optimization of Stereo-matching Algorithms Using Existing DEM Data

Here we present a new method for using existing Digital Elevation Model (DEM) data to optimize performance of stereo-matching algorithms for digital topographic determination. We show that existing DEM data, even those of a poor quality (precision, resolution) can be used as a means of training stereo-matching algorithms to generate higher quality DEM data. Existing data are used to identify and to remove cross surface errors. We test the method using true vertical aerial imagery for a UK upland study site. Results demonstrate a dramatic improvement in data quality even where DEM data derived from topographic maps are adopted. Comparison with other methods suggests that using existing DEM data improves error identification and correction significantly. Tests suggest that it is applicable to both archival and commissioned aerial imagery.