In: Wagner W., Szdkely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
. 
244 
< *-* 
Dead reckoning Image referenced 
georeferencing georeferencing 
Dead reckoning 
georeferencing 
Reference 
Figure 1. Georeferencing in GPS denied situation 
Satellite 
Resolution 
M 
Revisit time 
M 
Swath MdtlT 
|km] 
IKONOS (1999) 
0.82 
3~5 
11.3 
EROS-A (2000) 
1.8 
3-4 
14 
Quickbird (2001) 
0.61 
1-3.5 
16.5 
SPOT-5 (2002) 
2.5 
2-3 
60 
Orb View-3 (2003) 
1 
-3 
8 
FORMOSAT-2(2004) 
2 
1 
24 
CARTOSAT-1(2005) 
2.5 
5 
30 
ALOS PRISM (2006) 
2.5 
2-46 
35 
KOMPSAT-2 (2006) 
1 
4 
15 
EROS-B (2006) 
0.70 
3-4 
7 
WorldView-1 (2007) 
0.50 
4.6 (60cm) 
17.6 
CARTOSAT-2(2008) 
0.80 
4-5 
9.6 
GeoEye-1 (2008) 
0.41 
2.8 (50cm) 
15.2 
WorldView-2 (2009) 
0.46 
3.7 (52cm) 
17.6 
Table 1. Current high-resolution satellite imaging systems 
In a recent research for automatic georeferencing of airborne 
pushbroom scanner by Cariou and Chehdi (2008), the reference 
data is transformed into the acquired image domain using the 
initial EOPs from INS, and then mutual information is 
computed between the transformed reference and acquired 
image. Through iteration of this computation and image 
transformation, a pixel-to-pixel correspondence is obtained and 
used for estimating yaw angle, and small bias in roll, pitch and 
height constant. This approach requires good initial EOPs from 
the INS and is computation intensive as a large number of 
iterative image transformations are needed. 
This study proposes the combination of SURF (Speeded-Up 
Robust Features) (Bay et al., 2008) and RANSAC (Fischler and 
Bolles, 1981) for robust image matching, and the collinearity 
equation camera model with the Gauss-Markov stochastic error 
model for the trajectory modeling of the airborne pushbroom 
camera. The paper is structured as follows. First, the proposed 
method is presented, including a brief description on the image 
matching and platform trajectory modeling. Second, 
experimental results on simulation data are discussed, followed 
by a brief conclusion. 
2. HIGH RESOLUTION SATELLITE IMAGE AS 
GROUND CONTROL INFORMATION 
Since IKONOS-2 showed its potential in the commercial 
satellite image market, many high-resolution satellite imaging 
systems have been launched, see Table 1. The specification of 
high-resolution satellite images is listed in terms of its spatial, 
temporal resolution, and swath width. Note that many satellites 
provide sub-meter resolution with large swath width of more 
than 10 km. In addition, positioning accuracy has seen a steady 
increase over the years. For example, GeoEye-1 provides RPC 
with positional accuracy up to 2 m of circular error at a 90% 
confidence level (CE90) without GCP in the case of stereo 
images, and sub-meter accuracy could be achieved using a bias- 
compensation RFM model with a single GCP (Fraser and 
Ravanbakhsh, 2009). Moreover, higher performance satellites 
will be launched in the near future such as CARTOSAT-3, 
EROS-C, and GeoEye-2. These attractive capabilities motivate 
the idea of using high resolution satellite images as ground 
control information for other geospatial images, such as aerial 
images. In the navigation field, research has started on testing 
and suggesting the use of satellite imagery to support UAV 
navigation (Sim et al., 2002; Conte and Doherty, 2008). 
3. PROPOSED METHOD 
Figure 2 shows the flowchart of the proposed method. As direct 
georeferencing is unavailable, due to GPS denied condition, and 
the reference data becomes available, short duration of HSI and 
INS is processed for georeferencing purposes. Though the INS 
data is drifting without georeferencing fixes, it can still provide 
good approximation for georeferencing, and thus reference data 
is windowed with an error margin, so the image matching with 
the raw HSI is limited to a smaller reference area (subset). 
During image matching, SURF is utilized with RANSAC to 
mitigate the effect of mismatched points. Successful image 
matching provides ground control information for each extracted 
raw image point, and thus the trajectory and attitude are estimated 
based on this information. 
Figure 2. Flowchart of the proposed approach 
3.1 Subset ROI reference data 
Since images used as reference data tend to be large, it would 
take too much time and require a lot computer power if the 
whole reference image is used for image matching, fortunately, 
the INS-based estimation of location can provide good 
approximation to obtain a region of interest (ROI), using the 
inverse form of collinearity equation. The ground coordinate of 
ROI can be determined from Equation 1. The ground height 
information could be selected as a constant value from 
knowledge about the target area; note that error of the height 
will be compensated in the error margin terms.