GEOMETRIC AND RADIOMETRIC MODELS IN PROCESSING
SPOT IMAGERY FOR OBJECT-SPACE SURFACES
J. WU, M. J. Hsyu, C. H. Liu AND D. C. LIN
CENTER FOR SPACE AND REMOTE SENsING RESEARCH
NATIONAL CENTRAL UNIVERSITY
CHUNG-LI, TAIWAN, ROC, 32054
XVII ISPRS CONGRESS IN WASHINGTON, D.C.
COMM. III , AUGUST 2-14, 1992
ABSTRACT
For digital photogrammetry, an integrated approach to image matching and
3D positioning is proposed for stereoscopic SPOT images. The nonlinear
functional models for space resection and the integrated approach are ex-
plicitly written down. Specifically, we use both piecewise linear models for
time-dependent orientation parameters and a patchwise bilinear model for
height parameters at elementary ground resolutions.
Our experiments show that fewer than 10 ground control points are more
than enough to obtain in resection sub-pixel accuracies + 7.2m, + 4.6m and
+ 5.5m in a local horizon coordinate system, respectively. The integrated
approach can yield in one case a digital elevation model having + 5m as a
height root mean square error, when good approximations to terrain form
and large weight constraints on heights are prerequisites.
Based on our experiences up to now, we review at the end our models and
discuss their prospects.
Key Words: Space resection with SPOT imagery, Integrated image match-
ing and 3D positioning.
1. INTRODUCTION
For analogue stereoscopic images, conjugate im-
age points are best determined by an operator,
e.g., on comparators. Of course, when the num-
ber of image points grows, working loads on
the operator become very heavy. Thus, people
have endeavoured to take full advantage of elec-
tronic computers to facilitate image correspon-
dance processes since wide availability of digi-
tal images in early 1980's (e.g. Hannah, 1989).
And least squares image matching is a repre-
sentative algorithmic development (Ackermann,
1984). For digital stereo images, general prac-
tices follow a two-step solution by determining
conjugate image points first and then by inter-
secting directional imaging rays to arrive at 3D
point coordinates.
It is known that photogrammetric point posi-
645
tioning deals with geometric and radiometric dis-
tortions which result from taking optical pic-
tures at different times and camera stations. When
an analysis model takes perspective displacements,
relief displacements and radiometric degradations
all into account, the model becomes integral in
nature and is more suitable for photogrammetric
processing. Our proposed integrated approach
to image matching and 3D positioning has been
initially tested for aerial photos (Wu and Chang,
1990). Now the approach is applied to SPOT
stereo images with a model extension that al-
lows for time-varying characteristics of sensor's
orientation along orbital paths.
While our interests in research on this topic are
high, we feel motivated by engineers who desire
up-to-date wide-coverage accurate height data
for planning major governmental civil construc-
tions.