Radargrammetry is the technique for extracting three-dimensional information from radar images. It applies photogrammetric principles to synthetic aperture radar (SAR) images. By viewing an object from different positions separated by a baseline, the appeared object position will vary slightly (denoted parallax). The disparities for each position on the object are related to its x-y-z coordinates. In radargrammetry, such disparities are computed for an entire image. The result is the terrain elevation from the measured parallaxes between two (or more) images, acquired at different angles. Radargrammetry requires at least two SAR images acquired from different positions, normally across-track due to the configuration of a side-looking SAR. Same-side stereo-pairs with intersection angles in the range of about 10 – 20° have been a feasible compromise between reasonable geometric disparities and the accuracy of estimated heights. In general, the disparities can be estimated with higher accuracy as the angle of intersection increases (as the stereo exaggeration factor increases). However, the same points must be recognized in all images, and it is hence required that the images are as similar as possible. This improves the image matching and it is best achieved with small intersection angles, which furthermore decreases radiometric differences. A general procedure for generating an elevation model from stereo-pairs is applicable for radargrammetry when optical stereo images are replaced with the backscatter intensity of SAR images. One image is selected as reference and the other(s) is coarsely registered to the reference, e.g., by using the attached meta-data. The same points are then located in both images using image matching. A common matching criterion is the cross correlation coefficient. Then, spatial point intersections are computed, which is the least square approach to find the intersection points of SAR range circles as defined from the matched image pixels. The computed intersections result in a point cloud that finally is interpolated to a consistent elevation raster. The entire process is extensive and computationally expensive, and normally a dedicated software is required. Radargrammetry with images acquired from opposite sides have been little investigated, and was first limited to stereoscopic viewing. Some opposite-side research was later presented with limited outcomes under certain conditions. Most applications today will not consider opposite-side radargrammetry, since the alternatives are usually better. Same-side radargrammetry performs better than opposite-side, while interferometric SAR that is based on phase differences, may be even more accurate. One advantage of radargrammetry is however, that it remains less affected by atmospheric disturbances compared to interferometric SAR, because it is using the amplitude images.
Select and apply the radargrammetric equation
Completed