Multi-Resolution approach to Radargrammetric Digital Elevation Models Generation
Xiaojing Huang, Leong Keong Kwoh and Hock Lim
Centre for Remote Imaging, Sensing and Processing
National University of Singapore
SOC1 Level 2, Lower Kent Ridge Road, Singapore 119260
Fax: (65)7757717
Email:crshxj@nus.edu.sg
Key Words
Digital Elevation Models, Radargrammetry, Correlation
Abstract
In this paper, we present our work in Digital Elevation Models (DEM) generation with RADARSAT stereopairs, usually form from S2 and S6 standard beam modes, using radargrammetric technique. We introduced a weighted block window for the matching of corresponding patches and a hierarchical multi-resolution approach to improve the DEM generation robustness. Prior to DEM generation, an imaging model based on the satellite and target geometry, and signal acquisition time for RADARSAT has been developed, and the model parameters have been refined to match the local terrain with ground control points (GCPs).
1. Introduction
DEM of a region can be generated from stereopairs of satellite images. In tropical areas where cloud free optical images are hard to acquire, DEM generation with radargrammetry using high resolution spaceborne synthetic aperture radar (SAR) images, such as RADARSAT images, is a potential alternative. In this paper, we described the system we developed at CRISP for DEM generation from RADARSAT stereopairs. The main processing steps involved are (a) establishing the SAR imaging model and determining the parameters of the imaging models, (b) automatic DEM generation applying a weighted block window digital correlation and implementing a hierarchical multi-resolution approach.
SAR Imaging Model
In a SAR image, the location on the ground of an image sample (u, v) can be derived from knowledge of the sensor position and velocity [1]. The pixel coordinate u is related to the slant range r from the sensor to the corresponding point on the ground. While the line coordinate v is related to the azimuth time s at which the pulse of the corresponding line is transmitted. These relationships can be expressed as follows:
where r
0 is the slant range of the first pixel in the image, k
0 is the scaling factor in the range direction, s
0 is the azimuth time of the first line in the image, and ks
0 is the scaling factor in the azimuth direction. The preliminary values of r
0 , kr
0 , s
0 and ks
0 can be obtained from the RADARSAT product leader file.
The SAR range r and line time s of any target is further related to the sensor as follows:
where r
s, v
s are sensor position and velocity vectors, and r
t, v
t are target position and velocity vectors. f
dc is the Doppler
centroid frequency, and
l
is the radar wavelength. For RADARSAT SGF images, f
dc is zero since the image is resampled to the zero-Doppler. The target velocity can be determined from the target position since the target is on the earth surface. The sensor position and velocity can be derived from azimuth time s and satellite state vectors given in the image leader file.
Equations (1) - (4) describe the imaging model we use for a SAR image. All imaging parameters can be obtained from the image product leader file. To optimize the imaging model for the local area of interest, some parameters can be refined with GCPs. We choose to only refine equation (1) and (2) with the following expressions:
where
Dr and kr are the shift
and scaling factor of image slant range,
respectively,
D
s and ks are the shift and scaling factor of image azimuth time, respectively, r' and s' are the corrected slant range and azimuth time, respectively. There are thus a total of 4 parameters to be refined for each image. These 4 parameters are solved by least square adjustment techniques, using at least 2 GCPs.
To verify the correctness of the imaging model, we use them to terrain geocode a RADARSAT S2 image (fig 1) and a RADARSAT S7 image (fig 3) of Lantau Island (Hong Kong) with available DEM. The parameters of the imaging model for each image were refined using GCPs obtained from topographic maps. The good agreement of the terrain corrected geocoded images shown in figure 4 ( the S2 geocoded image shown in red and the S7 geocoded image shown in cyan) shows the correctness of the model.
Figure 1. Lantau Island in RADARSAT SGF image of S2 mode acquired on 25 Dec 1996
Figure 2. (below) Lantau Island location map.
Figure 3. Lantau Island in RADARSAT SGF image of S7 mode acquired on 28 Dec 1996.
Figure 4. Two images match very well after geocode to accurate topographic map (red for S7 mode, green and blue for S2 mode).
