Georegistration of synthetic aperture radar (SAR) images is an ongoing problem. We present a SAR-sensor independent method that enables assessment of image geolocation accuracy at nearly any terrestrial scene location. To achieve this, a collected image is co-registered against a global reference image set, such as the high-fidelity X-band RCS layer of the TanDEM-X High Resolution Elevation Data Exchange Program (TREx) data set, which has well-characterized horizontal and vertical errors. Measuring sensor geolocation accuracy against a high-fidelity broad-area reference set can mitigate the effect of location errors that, if left uncorrected, may have a deleterious impact on downstream applications. In particular, the proposed method could help automate the assessment and quality control of existing and emerging constellations of commercial satellite systems. Our co-registration approach first applies a pre-warp operation that exhausts known SICD (Sensor Independent Complex Data) metadata from a collected image to get within the neighborhood of the true geolocation in the reference image. Then we attain a fine scale registration through local feature detection and extraction, via an established computer vision algorithm. Finally, we compute a similarity transformation on matching points between the test and reference images; the transformation includes translation values, which are used for assessing geolocation accuracy, as well as scale and rotation estimates that are used for a confidence measure.
|