IU at the University of Utah: Extraction of Micro-Terrain Features

The construction of large-scale geospatial databases remains an expensive and time consuming task. We describe an approach for automatically extracting certain types of linear features with a horizontal extent significantly less than the resolution of the baselevel terrain data covering the area in which these structures occur. The first phase of this effort considers two classes of such features: ravines and road cuts and fills. Accurate detection and localization of these features is difficult in even high-resolution elevation data. While they are often apparent in aerial imagery, they are easily missed or confused with other features, making reliable detection based on imagery alone problematic. The key to solving this problem is to utilize techniques which combine photogrammetric analysis of the terrain with focused image understanding methods applied to individual images.