Evaluation of Digital Surface Models by Semi-Global Matching

Abstract

This paper considers a processing chain for automatically creating high resolution digital surface models and true ortho-images from aerial and satellite image data. It has been developed at the Institute of Robotics and Mechatronics of the German Aerospace Center (DLR-RM). The processing chain is based on Semi-Global Matching (SGM) that uses a radiometric robust matching cost and an optimization that is based on a global smoothness constraint. SGM is especially suitable for creating models of urban scenes, where sharp depth discontinuities and small details need to be precisely reconstructed. However, the technique also produces very good results in scenes with forest and mountains. In this paper we give an overview of the processing chain and evaluate its results on test data sets from different aerial cameras. It is concluded that SGM permits the creation of high quality surface models that are more accurate and provide much more detail than a surface model from an aerial laser scanner. We also discuss the conditions under which good surface models can be produced by SGM. For very good results, an overlap of 80 % or more along track and 70 % across track should be provided.