1117
UTILIZATION POTENTIAL OF HIGH RESOLUTION STEREO DATA FOR
EXTRACTING DEM AND TERRAIN PARAMETERS
P. Jayaprasad, Ritesh Agrawal, S. K. Pathan*
Space Applications Centre, Indian Space Research Organisation (ISRO), Ahmedabad-380 015,
India - (jayaprasadp, ritesh agrawal, subhan_kp)@sac.isro.gov.in
*Secretary Commission IV
KEY WORDS: CARTOSAT-1, Stereo data, RPCs, DGPS, DEM, Ortho Image, Accuracy Assessment, terrain parameter
ABSTRACT:
Cartosat-1 data provides along track stereo data continuously with its fore and aft cameras. 10 bit quantization and near real time
imaging between the stereo pairs which improve the image matching accuracies. Stereo data along with Rational Polynomial
Coefficients (RPCs), provides an opportunity for photogrammetric processing for DEM and ortho image generation. The present
study focuses on the effect of the number of Ground Control Points (GCPs) and polynomial order for the refinement of RPCs for
generating an accurate Digital Elevation Model (DEM). 10 GCPs were used in the refinement process and 9 Independent Check
Points (ICPs) used for accuracy analysis. Both GCPs and ICPs were derived using Differential GPS survey (DGPS). Effect of DEM
resolution while generating the Ortho Image is also studied. Terrain parameters such as slope, aspect and drainage network have been
automatically extracted. The drainage network was extracted at different DEM resolution using similar area specific (5000 m 2 )
accumulation threshold and a comparative analysis of the order of the drainages has been carried out. The accuracy analysis shows
that RMS errors of DEM and ortho images were within 5m both in elevation and planimetry respectively with 10 GCPs and
polynomial order 2.
1. INTRODUCTION
Launch of CARTOSAT-I on 5 th May 2005 has opened a new
era in civilian community for addressing newer applications at
larger scale. With its capability to acquire stereo images with
Fore (+26°) and Aft (-5°) cameras and a nominal B/H ratio of
0.62, it gives a stereo data at 2.5 m spatial resolution. Less than
a minute interval acquisition time between the stereo pair makes
an ideal imaging condition. The repetitivity of the satellite is
126 days with a revisit capability of 10 days and the scene size
of 30 km * 30 km has shown tremendous potential in the field
of satellite photogrammetry. Stereo pairs are formed either from
across track geometry or from along track geometry.
Advantageous and disadvantages of the along track and across
track stereo viewing mainly depends on the slope and aspect of
the terrain towards the viewing geometry and illumination
conditions of the two imageries. The Fore and Aft cameras
onboard provides a continuous strip of stereo data. The mono-
scopic camera on board IRS 1C/ID, Ikonos, SPOT and
QuickBird generates a stereo pair by steering the camera
(along/across track depending on the capabilities) and provides
a limited stereo coverage rather than a continuous one. For three
cameras system, in addition to fore and aft imagery, nadir
imagery also provides the satellite capable of taking triplet
imagery (ALOS-PRISM).
Satellite Photogrammetry techniques have been extensively
used by the scientific community in deriving high resolution
DEM, Ortho image and terrain parameters such as slope, aspect,
contours, drainage etc. Digital Elevation Model (DEM) has
become an inevitable component in most of the remote sensing
applications viz. infrastructure development, watershed
management and development, hydro-geomorphology, urban
morphology, disaster management etc. Keeping these
applications in view, the current study aimed at exploitation of
Cartosat-1 stereo data for various applications.
Rational functions models (RFMs) have gained popularity, with
the recent advent of high resolution data supplying Rational
Polynomial Coefficients (RPCs) along with stereo / mono data.
Providing these coefficients along with stereo data, instead of
delivering the interior and exterior orientation parameters and
other properties related to physical Sensor, one can proceed to
satellite photogrammetric processes which approximate the
sensor model itself.
A detailed study of the RFMs for photogrammetric processing
has been carried out by Tao and Hu (2001). Di et. al. (2003)
demonstrated different ways to improve the geo-positioning
accuracy of Ikonos stereo imagery by either refining the vendor
provided RF (Rational Function) coefficients, or refining the RF
derived ground coordinates. Poon et al. (2007) focuses on
Digital Surface Model (DSM) generation from high resolution
satellite imagery (HRSI) using different commercial of the shelf
(COTS) packages. They validated the stereo DEM with InSAR
DEM for different land forms. Nadeem et. al (2007) validated
DEM generated from Cartosat-1 stereo data. Crespi et. al. (2006)
evaluated the DSM by comparing the heights of several
buildings and points on the road axis derived from a large scale
(1:2000) 3D map. Fracer and Hanley (2005) demonstrated the
wide applicability of bias compensated RPCs for high accuracy
geo-positioning from stereo HRSI for a mountainous terrain.
Chen et. al. (2006) compared geometrical performance between
rigorous sensor model (RSM) and RFM in the sensor modeling
of FORMOSAT-2 satellite image. Dabrowski et. al. (2006)
evaluated DEMs generated with different numbers of GCPs
from Cartosat-1 stereo data at large number of evenly
distributed check points. Similar attempts to evaluate the
accuracy of the DEM using different number of GCPs have
been made by Michalis and Dowman (2006) and Rao (2006).
RFM based processing methods and mapping applications was
developed for 3D feature extraction, ortho rectification and