Istanbul 2004 
e (absolute 6 
as the ground 
omputational 
"ing sensors) 
attitude, and 
frame. All 
1 the ground- 
2 are imaged 
in the image 
1,2, and 4, 
position and 
y GPS-based 
e calibration 
GPS antenna 
ude (6 DOF) 
1s follows: 
(3) 
ection of the 
ormation). 
| orientation) 
‘or example, 
points in the 
ter” satellite 
? acquired by 
Wig, The 
er" exposure 
"master" and 
mmon plane. 
| established 
el 
y 
Y 85 
T La 
ction cosines 
the “master” 
ss 
X" satellite. 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B1. Istanbul 2004 
Because the three rotation angles in the *follower" satellite are 
the rotation from the “follower” image space to auxiliary 
coordinate system space, thus the relative orientation of the 
follower” satellite to the “master” satellite is (0. ; 9, ; ku): 
The relative position of two satellites can be determined by 
exposure center coordinates of two satellites, which have been 
described in Section 3.1.1. The final basic formation relative 
state is given by the following: 
Relative position [Bi, AB», AB», eeee, AB;,], and 
relative orientation 
G. Ao s 7=1-57=1-6 
0, ^; izlicSizlaó 
Kr ARE SG 
1,3? 
We have experimented the simultaneous determination of six 
absolute DOF and six relative DOF when satellites are various 
flying heights, in-track overlap, cross-track overlap and 
analyzed the influences factors to accuracy and reliability of 
absolute and relative navigations. 
3.2 Onboard “GCP” Recognition Based on GIS Data 
Formation fly of a satellite cluster requires the high accuracy 
determination of relative position and absolute position and 
attitude. Because of the low accuracy and reliability of 
navigation information by navigation sensor (Alonso ef al. 
1998, Gill et al. 2001), a few GCPs are necessary for highly 
accurate and reliable geocoding. An algorithm, which 
recognizes GCPs onboard via the support of a geo-database 
(GIS database), is now investigated. The steps in this algorithm 
will be presented next. 
3.2.1 Landmark Vector Data(base) Management System 
It is impossible to provide traditional photogrammetric target 
points at real-time during most satellite observing missions. 
We propose to use natural landmarks (e.g., a crossroad center) 
to replace the traditional GCPs; and we denote these landmark 
GCPs by LGCPs. They are stored on the onboard computer 
(Figure 2). The creation of an LGCP database includes 
landmark selection, data structure/model for 3D coordinates 
storage, reference frame datum (e.g., WGS$84), and datum 
transformation, fast query and retrieval algorithms. These 
algorithms and methods on ground have been implemented 
before and the details can be found in Zhou and Jezek (2000). 
3.2.2 Identification of LGCPs using Optical Correlator 
Eor onboard geocoding of remotely sensed satellite images, 
image coordinates corresponding to LGCPs must be known. 
The LGCPs stored in the onboard computer can be retrieved at 
real-time; thus, the core task is to precisely determine the pixel 
coordinate of corresponding LGCPs from onboard sensor 
images. The basic steps are: (1) create a template image of the 
LGCP from LGCP database; (2) determine the AOI (area of 
interest) in the sensor imagery via back-projection; and (3) 
match the template image and the sensor image via JTC for 
pixel coordinate determination of the LGCPs._ Next, these steps 
will be described in more detail. 
a. Creation of Template Image of LGCPs (Figure 4): 
The creation of a template image for LGCP is to convert 
landmark vector data into raster image data form. The gray 
values of vector landmark data in template images are assigned 
255, and the background is assigned 0. The size of template 
mainly depends up the texture content around the landmark, 
navigation errors, and image GSD (ground sample distance). 
301 
In our study, when the size of template is typically 50X50 
pixels? to 100X100 pixels”; GSD=Im, then the error of orbital 
position is 3~6 meters and the error of sensor attitude is 0.002 
degree. 
b. Area of Interest (AOI) Determination (Figure 4): In 
order to increase matching speed, we can narrow the search 
space of the match processing. By the “coarse” EOPs (position 
and attitude) provided by onboard navigation sensors and priori 
calibrated IO parameters, we can back-project 3D coordinates 
of LGCPs into the sensor image plane via Eq.l for the 
approximate location of landmark. Based on this approximate 
location, we can design an area of interest (AOI) in the sensor 
imagery. The size of AOI mainly depends upon the GSD, 
navigation error (other errors, e.g., atmospheric refraction, lens 
distortion, etc., are relatively less). A larger AOI increases the 
load of computation, and a smaller AOI cannot ensure 
sufficient search space. In fact, the prior EOPs and all 
distortions of the imaging system can be used to predict the 
search range. In our study, the AOI is determined using 200 by 
200 pixels”, because the offset between the ideal and actual 
positions of the same feature is about 18 m (about 18 pixels 
due to 1-m GSD), when the position and attitude error of the 
sensor are 3-6 m and 0.002°, respectively. Whatever the Earth 
observing satellite's specification is, the size of AOI should be 
ensured to provide sufficient search space for matching 
processing. 
c. Landmark Match using Optical Correlator: We will 
use Joint Transform Correlator (JTC) to realize the matching 
between the landmarks stored in the landmark database and the 
same landmarks in the sensor imagery, because the processing 
rate of the JTC can reach 300 pts/s (Janschek ef al. 1999, 2000). 
Moreover, the JTC can also work under some scale and 
perspective distortions between compared images. The 
matching procedure, when the JTC consists of one processor, is 
briefly described with the following steps. 
(1) During one cycle of matching, a temple image and an 
AOI image (the current image) from the sensor imagery 
are simultaneously entered into the optical system of the 
Optical Fourier processor (OFP) by a spatial light 
modulator (SLM). 
(2) In the focal plane of the lens, the image of the Joint 
Power Fourier Spectrum (JPS), which is detected by a 
square-law image sensor (usually CCD) and entered 
into the same SLM during read out of the CCD, is 
formed. Then, in the focal plane, the image of the 
correlation function is formed with two symmetric 
bright points - correlation peaks - if that current image 
contains even a part of the temple image. The position 
of the peak corresponds to the mutual shift of the 
current and temple images. The correlation image is 
processed by the digital processing unit (DPU) in order 
to detect the correlation peaks and calculate their 
position. 
Template image 
Landmark