Extracting a digital elevation model from epipolar pairs

Generating a digital elevation model (DEM) consists of the following steps:

Converting the raw images into epipolar pairs.
Epipolar images are stereo pairs that are reprojected so that the left and right images have a common orientation, and matching features between the images appear along a common x-axis.
Extracting DEMs from the overlap between the epipolar pairs.
The resulting DEMs are called epipolar DEMs. At this stage, they are not georeferenced.
Geocoding the epipolar DEMs and stitching them together to form a single DEM.
The result is one DEM reprojected to the ground coordinate system.

A DEM usually contains poorly correlated areas. You can correct these areas before or after each DEM is geocoded.

You can use either of the following two methods to extract the DEM:

NCC (Normalized cross-correlation)
SGM (Semi-global matching)

NCC (Normalized cross-correlation) produces lower-quality results with more errors and less detail, but with minimal processing time.

SGM (Semi-global matching) produces higher-quality results with fewer errors and higher detail, but processing time is increased greatly.

To extract a DEM from epipolar pairs

In the Stereo pair table, in the Select column, click the row of each epipolar pair you want to select or, to select all the epipolar pairs in the table, click Select All.
Under DEM Extraction Options, click the Extraction method list, and then do one of the following:
- To use NCC, make sure NCC (Normalized cross-correlation) is selected.
  That is, NCC (Normalized cross-correlation) is the default method.
- To use SGM, select SGM (Semi-global matching).
In the Output DEM vertical datum list, select a vertical reference for the elevation values:
- MSL: Elevations are referenced to mean sea level (MSL). The EGM2008 geoid model is used to convert the elevations to ellipsoidal heights.
- Ellipsoid: Elevations are referenced to the ellipsoid. The coordinate system of the DEM raster file is examined to determine which ellipsoid and datum to use. For example, if the DEM file is referenced to NAD83, the heights are assumed to be NAD83 ellipsoidal heights. If the DEM and math model are based on different coordinate systems, the heights are converted automatically to the math-model coordinate system before use.
In the Pixel sampling interval list, click the number of image pixels and lines (sampling frequency) to use to extract one DEM pixel.
If you are using RADARSAT data, do not use a sampling of one pixel, because it causes difficulty in correlating the speckle inherent in SAR data.

Note: The Pixel sampling interval list is unavailable when, under Geocoded DEM, the Create geocoded DEM check box is selected. Sampling is determined automatically based on the resolution of the output geocoded DEM.

For more information about pixel sampling, see Understanding DEM detail and pixel sampling.
In the Smoothing filter list, select if you want to apply any filtering for smoothing:
- None: no filtering is applied
- Fill holes only: fill holes only with no additional filtering
- Low: filter by using low smoothing
- Medium: filter by using medium smoothing
- High: filter by using high smoothing
To permanently remove the epipolar DEM from disk after the geocoded DEM is generated, select the Delete epipolar DEMs after use check box.
Removing the epipolar DEM after use also saves disk space.
To continue setting up automatic DEM extraction, do one of the following based on the extraction method you selected in step 2; that is, if you selected:
- NCC (Normalized cross-correlation), see Additional setup for NCC.
- SGM (Semi-global matching), see Additional setup for SGM.
If necessary, you can create a geocoded DEM and set up a start time for the extraction.
After you complete the setup of the automatic DEM extraction, click Extract DEM.
The extraction begins based on the specified start time.