Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.264265
Title: Automatic height extraction from stereoscopic SAR imagery
Author: Twu, Zway Gen
ISNI:       0000 0001 3540 1937
Awarding Body: University of London
Current Institution: University College London (University of London)
Date of Award: 1997
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Abstract:
This research was concerned with various aspects of deriving DEM information from ERS-1 SAR imagery. Stereoscopic SAR offers the possibility to determine a DEM of large areas and is complementary to interferometric SAR (IFSAR). Compared with former SAR image data sources, ERS-1 has the advantage of providing accurate ephemeris data. In addition, it has different image modes which enables the verification of results from various research groups regarding geometric configurations. One of the aims of this study was to establish a standard model for deriving DEM accuracy, based on ERS-1 as the image data source, that would be applicable for all other new types of SAR imageries. An example of such SAR imageries worth investigating in the future is RADARSAT, which provides versatile image modes with a wide range of incident angle and resolution. In summary, the studies carried out for this research project could be described under two subheadings, namely matching and intersection. With respect to the matching, pyramidal stereo matching techniques were used in combination with an excellent area-based region growing algorithm to achieve dense coverage. The special interest in this study is the initial seed points used for the pyramidal matching process employed were chosen randomly instead of by manual selection. To examine the function of these random seed points, the original matching algorithm was modified to have four extra values in its output, which were later found to be able to aid the determination of the advantages of using image pyramids. It was also discovered that the disparity sum was a good measure for judging the factors affecting the matching accuracy in most of the studies. As a result, this parameter was used to investigate different strategies of pyramidal matching, to pin-point the link between the upper and lower tier in the image pyramid, as well as for the removal of the blunders. The analytic method was used in this study for the intersection procedures and based on this, two new approaches were developed. Both of these approaches were found to significantly enhance the DEM accuracy and their success was determined to be due to the linear-correspondence relationship between the image and the object space. Another important discovery was that any alteration made on the slant range of matched image pixels would only result in changes of terrain height values. Different geometric conditions for the three pairs under study were also analysed. It was concluded that the convergence angle of two given orbits would have considerable influence on the intersection accuracy, the smaller the angle the greater this influence. It is demonstrated in this thesis that with a same side convergence angle of 2.14°, the intersection error could reach 426.95m for one Y pixel shift. The above phenomenon explains the underlying reason why the DEM accuracy could not be improved to the same accuracy as, for example, SPOT data. To summarise, a satisfactory DEM could be obtained from ERS-1 images using the approaches developed in this study which could reach an accuracy of 20.18m for the same side and 12.23m for the opposite side with the coverage of better than 80%. However, the orbit information unique to ERS-1 was observed to play an important role in the accuracy of DEM derived using the methods developed. If this information was not provided, other rigorous alternatives are required for its determination and these were investigated. This research project has made its contribution by establishing a general model able to determine the factors that would influence the accuracy of the pyramidal matching on the SAR imagery, as well as the development of different approaches based on the object domain to greatly increase the DEM accuracy. Altogether, the results obtained in this study should be a valuable source of information for other similar work to be carried out in the future.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.264265  DOI: Not available
Keywords: Pattern recognition & image processing
Share: