International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
length of camera.
exterior element
orientation image
offset curve
y observation curve
offset
t
Figure 2 Lagrange offset correction curve
Considering the GSP offset, IMU alignment and drift errors the
main factors to influence the accuracy of the exterior orientation
(Xy, Yu, Zu, Ow. Qu, Kx) of the N" image line and taking them as
unknown parameters in triangulation we obtain the following
model as:
Xw 7 Zaps * Zofiset
Yw 7 ZGps + Zoffset
Zw 7 Zaps * Zottset
ON = OMu Ÿ Walignment + Orit ex
PN = PIMU iF Palignment + Paritt ® t
Kw — KiMU + Kalignment + Karin © t
where, (Xaps, Yaps, Zaps) are the observation values of GPS
antenna in mapping coordinate system; (@mu, mus Kımu) are
the observation values of IMU in mapping coordinate system;
(X oilet - Y offen. -Z ote). Ae correction values for GPS offsets;
(Datignments alignment Kalignment) are correction values for IMU
alignments; and (gritts Parifis Karin) are correction values for IMU
drift errors with time t.
Combining Equation 1 and 2 we can get the extended collinear
equation integrating the GSP offset, IMU alignment and drift
errors for bundle adjustment as conventional aerial
triangulation.
3 Test Flight and Experimental Results
In order to check the performance of the camera, the stabilizer,
GPS/INS integrated unit, synchronization system, and
georeferecing software package, a test site has been established
in Yoriityuu, Saitama with more than 120 ground control points
over 50 km”. In December 2000 three 10km parallel flight strips
with two 3km cross flight strips of TLS imagery were obtained
with STARIMAGER over the test area (Figure 3). More than
30% side overlaps between the three parallel strips are ensured.
The following parts will give the test results for different block
configurations, different number GCPs used in triangulation and
different section lengths of strips.
3.1 Single Strip
One obvious character of TLS is a long continues stereo image
strip of ground linear objects such as roads, railways, rivers and
shore lines etc. could be collected. Single strips are frequently
obtained. To ensure geo-referencing accuracy these single strips
with reasonable number of ground controls, we tested these
strategies:
e Different section number
9 Different GCP number
€ Different GCP configuration
From the 5 single strips we use the middle strip of the three
parallel ones as sample to report our tested results based on these
strategies. In this strip 30 GCPs and 578 pass points were
measured with semi-auto matching method over 10 km length
and 0.6 km width.
1) Different section number
Using Equation 2 to compensate GPS/IMU observed data errors
for a long strip TLS imagery cannot produce satisfactory results
due to some non- systematic or local distortions caused by
turbulence of platform. To alleviate this problem we designed a
continues subsection approach in the following characters:
€ It localizes, and thus simplifies, the complex distortion
by logically dividing the entire image strips into a
minimal number of logical pieces (Figure 4). Such a
division is intuitive and easy to implement. The
simplified distortion of each piece can then be modeled
using bivariate polynomials with a high accuracy level.
€ Since the division is not physical, the continuity of the
whole image strip can be retained using the concept of
anchor GCPs and pass points. In other words, the whole
image strip, after being assembled from the pieces, will
form a seamless strip within the framework of the
chosen coordinate system.
Figure 3. Layout of TLS imagery for triangulation test.
38
International Archive
I
I
Anchor GCP/ P:
Ist logical strip
Figure 4. Thec
SI
Section RN
Length
20000
30000
40000
50000
100000
Whole
Table 1. Statistics
From this table we
accurate the results.
2) Different GCP
To investigate the ii
we tested several c:
number. Table 2 li
number for 30000 li
GCP RN
Number
6 GCP
8 GCP
Il GCP
12 GCP
All GCP
Table 2. Statisti
