International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004 
  
A Bundle Block Adjustment was carried out in the above 
conditions using GeoTex software (Colomina et al. 1992). 
Focal length and the principal point parameters were adjusted. 
At a first stage, no corrections for lens distortion were taken 
into account. 
In a second step, a polynomial of S order was adjusted in order 
to remove the lens radial distortion. Figure 5 shows the radial 
component of the photogrammetric residuals against the 
distance to the principal point and adjusted polynomials for the 
right and left cameras. 
LEFT CAMERA RIGHT CAMERA 
  
adjusted polynomial adiusted potyremie 
Radial residuals in. pixels 
  
Radial residuals in pixels 
  
  
4 
3 
2 
1 
i. 
0; 
1 
2 
3 
4 
Ro! M c © m oc a 
  
  
  
i 
i 
; res i ; 
€ 1000 2000 3000 4000 5000 6000 7X0 8000 € 1000 2000 3000 4000 5000 6000 7600 
Radus in microns Radius in microns 
Figure 6: radial distortion calibration: adjusted polynomials for 
lens radial distortions for right and left cameras. 
Notice that no significant differences were found between the 
radial lens distortion of each camera. As a conclusion, for this 
set of optics and cameras a single polynomial can be used 
independently of the camera and lens to model lens radial 
distortion. However, the calibration parameters are not 
interchangeable because focal length and principal point 
parameters are significantly different between each camera. 
3.2 Camera boresight calibration 
In order to be able to compute the absolute position of any 
photograph in the object space, it is mandatory to compute 
accurately the relative position and attitude of each camera to 
the inertial reference frame defined by the GEOMOBIL 
orientation subsystem. 
The calibration site is in the neighborhood of the ICC and 
consists of two cylindrical walls in an open environment (with 
excellent GPS visibility). On these walls, about 60 points are 
surveyed with an accuracy of 1-2 cm. A GPS Ground Reference 
Station is set close to the calibration site. 
In the procedure, the wall is imaged by the GEOMOBIL system 
from different positions, azimuths and distances. A few 
stereopairs are selected from this set of images. The selection 
criterion is to obtain some stereopairs at different distances, 
azimuths and positions of the GEOMOBIL with respect to the 
calibration site. The acquisitions are performed in static and 
dynamic mode (van in movement) Dynamic acquisitions 
demonstrate that the synchronization subsystems work as 
expected. 
Wall control points are identified in the selected images and a 
Bundle Block Adjustment is performed. In the adjustment, the 
adjusted camera calibration parameters are taken into account 
(focal length, principal point and lens distortion). The goal of 
the Bundle Block Adjustment is to determine a set of boresight 
parameters per camera (eccentricity vectors and misalignment 
matrix between the image reference frame and the inertial 
reference frame) and a set of relative orientation parameters 
(camera relative orientation). 
266 
Adjusted relative orientation obtained accuracies of 1 cm for 
position and 60-80 arc seconds for attitude. Adjusted boresight 
parameters obtained accuracies of 1-2 cm for the eccentricity 
vector and 120-150 arc seconds for the misalignment matrices. 
            
      
Figure 7: À stereopair of a calibration data set with some 
control points identified and marked on the images. 
No significant residuals in the position and attitude parameters 
(orientation) of the dynamic acquisitions were found. Thus, it 
may be concluded that the synchronization subsystem has 
neither drift nor biases that affect image timetagging. 
Once the boresight parameters are computed, the GPS/IMU 
subsystem orientation parameters may be transferred to the 
images. Preliminary results on the empirical accuracy of the 
system using direct orientation are summarized in table 2. Up to 
39 objects in the calibration test field were identified in 
photogrammetric models when the van was moving (at 16-18 
meters distance of the wall) and its coordinates computed using 
direct orientation techniques. The coordinates were compared to 
the coordinates computed using surveying methods. As the 
azimuth during acquisition was nearly zero degrees, northing is 
approximately along-track and easting and H are across-track. 
Note that these empirical accuracies are coherent with the 
theoretic accuracies shown in figure 3. 
  
  
  
  
  
  
  
o 
Easting (across-track) 0.05 m 
Northing (along-track) 0.13m 
H (across-track) 0.03 m 
  
Table 2: Empirical accuracies 
4. PRACTICAL RESULTS 
Some missions have been carried out under different 
environmental conditions. In this article, we focus on the results 
obtained by one of them. The mission under discussion is an 
urban case. 
4.1 Mission results 
The acquisition took place in Sitges, a tourist resort town near 
Barcelona, on 5" November 2003. The acquisition was 
performed with the GEOMOBIL image subsystem and the 
recently integrated terrestrial LiDAR. A GPS reference station 
was set within a 5-10 Km distance from the mission site. 
In Sitges there is 1:1000 3D digital cartography, which was 
used to check the GEOMOBIL accuracy and precision. Map 
accuracy is 20 cm (1.64 6) per component. 
From the whole GPS/IMU trajectory, only two pieces with 
excellent GPS visibility have been taken into account. These 
pieces are referenced as zone 1 and zone 2 in the following 
lines. As some discrepancies between terrain and trajectory had 
been detected in the original GPS/IMU trajectory, a new one 
   
Inter 
  
has | 
new 
In s 
som 
extre 
flow 
zone 
com, 
Sitge 
  
  
  
  
  
Noti 
Prec 
heig 
Nort 
Diff 
traje 
sum 
is gc 
dete 
5.1 
Lasc 
each 
conf 
dire 
calit 
It is 
CCI 
calit 
wert 
posi 
poin 
Bun 
lasei 
coor 
and 
adju 
mis: 
para 
refei 
lasei 
Som 
miss 
Furt 
intes