Stereo processor

Stereo Processor is an integrated software package specially intended for generation of the ground relief digital elevation models (DEM) via processing of the pairs of images acquired by spaceborne SAR's.

Stereo processor provides for users an extraction of terrain height information from stereo pairs of SAR satellite imagery. It allows to generate an accurate digital elevation models (DEMs) of imaged ground surface.

The inputs of Stereo Processor are RADARSAT imagery distributed as georeferenced imagery of SGF or SGX data types. All input data should be in CEOS format. Sophisticated radar sensor modeling ensures the precise DEM generation results. The needed parameters for processing are derived by importer automatically from the CEOS headers. It is done simultaneously with loading of SAR imagery into internal environment.

Co-registration of the stereo pair of images could be done by two ways: through the tie ground point and using platform ephemeris data. Stereo processor correlates stereo pairs of SAR imagery automatically to extract radar parallaxes needed for height extraction. User may define and customize correlator in order to meet the local ground features. But the correlator can work without user's participation.

The resulted DEM pixel spacing is user's definable. An output DEM accuracy could be achieved up to several metes that mostly depend of input data spatial resolution.

The output of processor is DEM in geographic projection on WGS-84 ellipsoid.

Review of processing flow

Import and auxiliary data handling Reading of SAR data files. Reading of CEOS data files. Generation of parameter’s set needed for processing.
Georeferencing (for slant range data only) Transformation from slant range to ground range. Re-sampling to grid with same X and Y axes.
Co-registration of images Rotation and re-sampling of Slave image relatively to Master one with use of Tie Points. Selection of overlapping areas on both images.		Overlapped stereo images with same sizes
Pre-processing Subset of interested areas. Low-pass filtering.
Parallaxes to heights conversion rate Calculation of coefficients for parallaxes-to-heights conversion from CEOS ephemeris. Calculation of coefficients for heights conversion with use of Tie Points.		Parallaxes-to-heights coefficients
Stereo matching Generation of radar parallaxes matrix. Generation of correlation functions matrix.		Radar parallaxes matrix Correlation values matrix
Smoothing Low-pass filtering of parallaxes matrix.		Smoothed parallaxes matrix
Relief heights generation Parallaxes to reference heights conversion. Absolute heights calculation from GCP. Absolute heights calculation from ephemeris data.		Heights matrix on the SAR path projection grid
Transformation to reference projection Geocoding of heights matrix. Orthorectification of heights matrix.		Absolute heights matrix on the cartographic grid with user's determined pixels space

Approaches to stereo images coregistration

Coregistration step is needed for proper overlapping of two stereo images as the preparation for the followed joint processing. Coregistration includes turning of one of stereo pair image relatively to other and coregistration of images along the SAR platform path (azimuth) axes.
Below two coregistration approaches are regarded which are both realized in current version of Stereo Processor software:
coregistration with use of two pair of tie points
coregistration via surveys ephemeris.

The coregistration approach with use of two pair of tie points may be described as following. The user set two tie points on each image of stereo pair. This operation is conducted interactively, i.e. both images are opened into separate viewers, and user did this operation through GCP tool. Each tie point on one image corresponds to tie point on other one. The aim of the turning of slave image relatively of master one is their overlapping in such way that the azimuth file coordinates of same ground elements became corresponded on both images with some constant value. After that this constant value could be transformed to zero one via shift of turned slave image along the azimuth exes. All these successive operations allow reduce the matching procedure to one-dimensional case. The one-dimensional matching is more preferred than two-dimensional one because of much less computational cost and time of processing. However, in principal it could be achieved only if all imaged ground surface elements are located on the same elevation level. For the ground surface, which shows some expressed relief, that is a common case, the precise overlapping for all surface elements could not be achieved. This fact leads to necessity of two-dimensional matching procedure.

Coregistration on an ephemeris. As for the previous approach, the two tie points should be set on each scene of stereo pair. The main difference of second coregistration approach from first consists of those features, that the turning angle are calculated upon the some reference surface independently of tie points location and the elevation of reference surface may be controlled by user. For resolving of this task we will use an ephemeris of each surveys, which could be derived from auxiliary date accompanying processed scenes. These allow us to realize the procedure of the chose of reference surface in such way that minimizes the coregistration errors induced by different elevations of the imaged relief.

This is a 3-D presentation of Canadian west coast site. Picture is being composed through draping of Radarsat SAR image (standard mode) over a relief digital matrix generated via Radarsat SAR stereo pair processing.