Stereo-assisted SAR Interferometry
Contact Persons
Keywords
Stereogrammetry, Interferometry, SAR, Digital Surface Model, DSM
Quick Reference
The two most reliable methods for extracting surface topography from SAR image pairs are interferometry and stereogrammetry. Although the high-resolution results obtained by interferometry have been the main focus for research into digital surface model (DSM) generation in recent years, it has been shown that the use of a lower- resolution DSM, obtained in this study by processing a SAR stereo pair, can aid in the generation of an interferometric DSM. Because of the increasing availabilty of stereoscopic and interferometric coverage generated by air- and spaceborne sensors in the near future, the combination of stereo SAR with interferometry will become increasingly feasible.
Because of its inherent higher accuracy, we choose to generate the DSM using interferometry, while using the previously-obtained stereo DSM to assist and automate the process. The stereo technique can be used to aid interferometric DSM-estimation in three independent ways:
- The stereo DSM can be used as a "first guess" for the interferometric DSM-estimation process. In practice, the stereo DSM is used to simplify (flatten) the interferogram before its conversion to a height model, reducing the chances of errors.
- A set of "ground control points" (GCPs) is automatically selected from the stereo DSM (using a threshold on the expected reliability), and used to calibrate the interferometric phase. This makes it possible to convert the phase to topographic variation in a pre-defined reference system.
- Any remaining holes, due for example to low interferometric coherence, are partially filled using stereo heights of expected high quality.
Finally, the smallest of the remaining DSM holes are filled in using cubic interpolation, taking into account the values at the hole boundaries.
The goal of this technique is to combine the robustness of stereo SAR with the precision of InSAR.
Example of Results
Stereo-assisted InSAR requires four acquisitions in general: two suitable for interferometry, two for stereo. Naturally, only the area common to all acquisitions results in a combined DSM.
Near-simultneous interferometric and stereo pairs were acquired using an airborne X-band sensor (courtesy of AeroSensing, now Intermap Technologies) over a test site north of Berne, Switzerland. The area, with a total height variation of 135 m, includes rolling hills, forest stands, and an artificial canal. The sensor was flown from east to west, or right to left in the results shown below. The terrain was illuminated from the south (bottom), creating shadow zones north of the forest stands.
An extract of a 1:25000 topographic map (courtesty of SwissTopo) for the test site is shown below, as well as the DoSAR reference DSM used for validation (please refer to the publications for details on this height model). The site is approximately 2x1 km in size.
All height models are shown artificially illuminated from the north-west.
1:25000 topographic map extract:
DSMs were estimated using stereo only, InSAR only, and our combined technique.
The estimated DSMs are shown below, along with the associated height errors in each case. Colours represent differences with respect to the height reference, with green representing no difference (key below the images).
The stereo DSM in (a) has mainly two large sources of height error: strong over-estimation in and near the shadow zones (red), and strong under-estimation of vegetated surfaces facing the sensor (blue). However, the scene is free of holes and quite accurate in the remaining parts of the scene, including the forest stand upper surfaces.
While few stark errors can be seen in the pure-InSAR DSM, the volume scattering within the trees, causing low phase coherence, made reconstruction of the forest stands impossible. Elsewhere, the flat-to-rolling topography could be easily be interpreted by the InSAR technique. Indeed, the visual appearance of the reconstructed terrain in (b) as compared to (a) is much more pleasing, and fine details such as the canal and single trees are just resolved using InSAR.
Combining the robustness of stereo with the fine resolution of InSAR results in the DSM shown in (c). Note that this DSM has been generated without any manual intervention. That is, whereas GCPs had to be hand-selected from maps in order to calibrate the phase in (b), phase calibration was done automatically in (c), using a large number of GCPs selected automatically from the result in (a). These points were not chosen at random; a number of criteria described in the publications had to met for a given point to be used in the final DSM.
The improvement offered by the stereo-assisted technique is clear in the last result. With more stable phase-interpretation made possible by the addition of the stereo DSM, the height errors are restricted to small areas where bad stereo heights were used. While most of the extreme stereo height errors were correctly avoided during the merging, some will inevitably "leak" into the final DSM. Nevertheless, the combination of both techniques produces a DSM much more similar to the reference than can be obtained by InSAR or stereo alone.
Methods
Please refer to Publications.
Partnerships
Research has been conducted in cooperation with the following organizations:
- Intermap Technolgies (AeroSensing)
- German Aerospace Centre (DLR)
Publications
- Extraction of surface topography from SAR images using combined interferometry and stereogrammetry (EUSAR 2004)
- Robustness of wavelet-based stereo matching for variable acquisition geometries using simulated SAR images (IGARSS 2002)