Urban area operations have become extremely important in the post 9-11 world, and tactical operations in urban areas require high-resolution digital elevation information. High-resolution DEMs are useful for line-of-sight analyses, mobility studies, precision targeting, and vector feature extraction and attribution.

Recent advances in the synthesis of Global Positioning System (GPS) and Inertial Measurement Unit (IMU) data allow for the precise computation of the location and orientation of airborne sensor platforms. This, in turn, provides the framework for automatically computing POSition Exterior Orientation (POSEO) solutions necessary for automated stereo correlation (ASC) processing of digital image data for extraction of high-resolution DEMs in urban areas.

We are conducting research to assess the resolution and accuracy of high resolution DEMs obtained from ASC processing of airborne digital imagery. In previous research, we demonstrated that urban-area DEMs with 3-m horizontal (xy) resolution and 2-m RMS vertical (z) accuracy could be produced from ASC processing of 1-m resolution black/white NAPP photography acquired by the USGS.

We are extending this work to include 0.3-m resolution digital color (RGB) imagery being acquired for homeland security needs under the joint NGA/USGS 133 urban cities initiative. The example below shows a sample image of an interstate interchange in St. Louis, MO with mixed residential and commercial land use. ASC processing of the imagery was used to produce a 3-m xy resolution DEM shown next to the image. The DEM is shown in raster format using a gray-scale representation where low elevations appear dark and high elevations appear bright. Single-family residential houses, commercial buildings, and elevated parts of the highway interchange are clearly resolved in the high-resolution DEM. The vertical accuracy of the DEM was assessed using several thousand check points derived from precision kinematic differential GPS survey. The RMS vertical accuracy of the DEM was determined to be 0.8 m.

The spatial resolution and vertical accuracy of the DEM extracted from the NGA/USGS is almost as good as DEMs generated from airborne Lidar surveys. The advantage of this approach is that the DEM is extracted from existing digital imagery and this is a cost-effective alternative to expensive airborne lidar surveys.

Once the high-resolution DEMs are extracted, automated feature extraction techniques can be applied to extract vector features. This is particularly useful for automated extraction of 2D building footprint vectors and 3D building shape features. An example of 2D and 3D building extractions from a high-resolution DEM are shown below.

We have applied these building extraction techniques to large datasets and tested the accuracy of the extractions. The figure below shows roughly 70,000 2D building polygons extracted for the city of Springfield, MO. Rigorous multi-statistic evaluations of the building extractions shown in Table I indicate that detection rate varies from 94-100% for residential vs. commercial areas while achieving correctness rates from 86-97%.