[ON 
>nsor data. 
quire both 
/ for direct 
metric and 
nnels with 
mance for 
3 software 
compared 
aluation of 
f different 
|! manually 
he Optical 
only the 
003b) and 
1atic DSM 
C aspects. 
ements in 
ition have 
, although 
oying line 
ims used 
jagery are 
rformance 
ET digital 
it forward 
veloped at 
this point 
employs a 
malysis is 
the DSM 
'aluated in 
. manually 
> has been 
already in 
allel CCD 
n — three 
red, green 
n addition, 
janix) for 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004 
In this study, datasets have been acquired with two different 
sensor plate (SP) designs. The standard and most known 
configuration applies to SPI design with staggered 
panchromatic lines in backward (PANB), forward (PANF) and 
nadir (PANN) position, the multispectral red, green and blue 
lines (RGB) between forward and nadir lines and the near 
infrared line (NIR) close to nadir. In the SP2 design, the RGB 
triplet is set at the nadir position (preferred for true orthophoto 
generation), two panchromatic lines in forward viewing 
(PANF), one panchromatic line for backward viewing (PANB) 
and one near infrared line between nadir and forward. In SP2, 
the staggered mode for the PAN CCDs is not used. In Figure 1, 
the SP1 and SP2 designs are illustrated and in Table I, the 
respective viewing angles of the CCDs are listed. 
UL AME 
/ 
7 7 PANN NIR PANB PANF PANF NIR = N 
(a) (b) 
Figure 1. Configuration of the CCD lines on the two 
different sensor plate designs, SP1 (a) and SP2 (b). 
The panchromatic lines forward, nadir and 
backward are indicated as PANF, PANN and 
PANB respectively, near infrared as NIR and the 
color triplet as RGB. 
   
  
CCD lines | SPI | SP2 
PAN 3289 09. -]4? +28°, +16°, -14° 
RGB +16° 0° 
NIR 2 +14° 
Table 1. Viewing angles of the CCDs for the two different 
focal plate designs. 
The acquired ADS40 channels are further rectified onto a height 
plane (Levl images) in order to be used for stereo viewing and 
in automatic matching processes for tie point and DTM/DSM 
extraction. In Levl images, differences between channels due to 
scale, but also rotation and shear that exist in raw images (LevO 
images) are removed to a large extent. However, the option of 
utilizing LevO images in aerial triangulation (AT) is being 
investigated (not handled in this paper), in order to accelerate a 
part of the ground processing chain (rectification and automatic 
tie point extraction). 
2. DATASETS AND SYSTEMS 
2.1 Datasets 
Two ADS40 datasets were used in these investigations. The first 
was acquired over the rural area of Waldkirch area in 
Switzerland and the second over the dense city center of 
Yokohama in Japan. The Waldkirch block was flown with SPI 
camera by LGGM in May 2002, and consisted of 4 parallel and 
2 cross-strips. The coordinate system used was the WGS84. The 
block of Yokohama consisted of three parallel strips and was 
403 
flown by Pasco Corp. The coordinate system used was the 
Japanese grid. Both datasets included panchromatic and 
multispectral imagery in LevO (raw) and Lev1 product with 0.20 
m ground sampling distance. 
In terms of radiometric quality, the Yokohama compared to the 
Waldkirch dataset exhibited higher noise. Interpretability of 
objects was more difficult, as denser and higher buildings 
existed in combination with strong shadows (in many cases 
saturated) and poorer radiometric quality (Fig. 2). All images 
used for DSM extraction have been pre-processed in order to 
reduce noise, improve feature definition and minimize 
radiometric differences among channels (Pateraki and 
Baltsavias, 2003a). This part was essential, notably for the 
Yokohama dataset, in order to help matching in shadowed areas 
(Fig. 3). 
Regarding geometric quality, each individual camera with SP1 
and SP2 has been calibrated over a test field with precisely 
measured control and check points, and interior orientation and 
IMU misalignment parameters have been estimated. These have 
been later used in AT which was carried out for both datasets, 
using ORIMA software, in order to adjust and refine the 
orientation parameters acquired from the GPS/INS systems on 
board. Tie points were automatically measured using Automatic 
Point Matching (APM) module of SS software. Blunders could 
be visually controlled and iteratively eliminated. GCPs in the 
Waldkirch dataset were distributed at the block corners, whereas 
for Yokohama at the block center. The derived geometric 
accuracy from bundle adjustment in terms of sigma a-posteriori 
was 2.5 um for the Waldkirch dataset. For the Yokohama 
dataset the geometric accuracy was lower, due to the poorer 
radiometric quality (more blunders in automatic tie point 
measurement), small errors in the recordings of the GPS and the 
poorer block geometry (no cross strips). Table 2 summarizes 
acquisition and bundle adjustment parameters of the two 
datasets. 
  
Figure 2. Original image quality of Waldkirch (top) and 
Yokohama (bottom) dataset.