High-resolution photogrammetric surface extraction over glaciated regions from WorldView stereo pairs
Abstract
The monitoring of surface change in glaciated regions such as Alaska, Greenland and Antarctica is an important pursuit in climate-related Earth Science. Repeat Digital Elevation Models (DEM) created by photogrammetric surface extraction from a time-series of stereo pairs provide an efficient and low cost means for analyzing surface change over large, remote areas. Stereo-photogrammetric DEM extraction over glaciated regions is challenging due to typically low-contrast surfaces such as ice, snow, mountain shadows and steep slopes, resulting in large feature search areas and matching failures. A method for reducing the feature search area is critical for successful and efficient DEM extraction in this terrain. The SETSM (Surface Extraction with TIN-based Search-space Minimization) algorithm is developed for overcoming these problems and performs surface extraction automatically, without any user-defined or a-priori information, such as seed DEMs, using only the sensor Rational Polynomial Coefficients (RPCs) for geometric constraints. Rotation-invariant, multi-patch Normalized Cross Correlation (NCC) is used as its basic similarity measurement. SETSM constructs a TIN (Triangular Irregular Network) in the object-space domain in order to minimize the necessary search space. It employs a pyramiding strategy that uses iteratively finer resolution TIN's to minimize the search space and uses a vertical line locus to provide precise geometric constraints for reducing the search area. As a major benefit, SETSM relatively adjusts the Rational Function Model (RFM) between stereo pairs to reduce the offset between corresponding points projected by the vertical line locus caused by RPC errors, dramatically reducing the number of matching failures. In SETSM, this offset is iteratively removed with a parabolic adjustment of the NCC solution. As a demonstration, Worldview stereo pairs for a variety of test areas in Alaska, Greenland and Antarctica are selected for creating 2m grid-spacing DEMs and validating the SETSM algorithm. Each DEM granule was mosaicked to a master grid by a simple height co-registration and split into 20 x 20 km tiles. Qualitatively, most surfaces in snow and ice-covered areas, mountain shadows and steep slopes are exactly reconstructed by SETSM. In addition, SETSM resolves edges and height of discrete features, such as icebergs, crevasses, rocks and streams at high detail. For analyzing DEM accuracy quantitatively, Operation IceBridge data is utilized for validating the height accuracy in Kangerlussuaq, Greenland. The RMSE mean and standard deviation without co-registration are 1.92 m and 0.33 m, respectively. 3D view
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2013
- Bibcode:
- 2013AGUFMEP53A0788N
- Keywords:
-
- 0758 CRYOSPHERE Remote sensing;
- 1926 INFORMATICS Geospatial;
- 1240 GEODESY AND GRAVITY Satellite geodesy: results