4.2 Single Stereopair Processing 
The camera parameters resulting from the self-calibration are 
shown in Table 1. As the image units used are in pixels and 
considering the pixel size is 8.75 u m, then it’s easy to see that 
the calibrated camera interior parameters are very closed to 
those mentioned in section 2.1. 
Table 1 Camera Parameters 
  
  
  
  
  
  
  
Para. Self-calibration 
camera 1 camera 2 
fx 916.885 920.403 
fy 916.885 920.403 
x0 326.939 325.119 
y0 247.464 239.608 
K1 1.829¢-7 1.959e-7 
  
  
  
  
  
The relationship between the camera center and GPS antenna 
can be determined through distances as shown in Table 2. Only 
distance relationships are constant as the bar is moved, rotated 
or tilted during the image acquisition. For the purposes of this 
experiment, the camera/GPS mounting bar is considered as a 
rigid body(even though because the reason of material used, the 
bar was still distorted during photography). For further dynamic 
mapping purposes, additional sensors for tilt orientation of the 
bar maybe needed. 
In this experiment, the distance relationship between the 
cameras and the GPS antennas in fact are preserved, but the 
calibration results and the GPS surveyed results of the three 
stereopairs are quite different from each other. This is just a 
result of the image coordinate measurement, GPS observations 
and the limited control accuracy. 
Table 2 Relationship between GPS and camera (m) 
  
  
  
  
  
  
  
  
parameter station] station2 station3 
| DGleftCleft 0.139 0.122 0.139 
DGrightCright | 0.093 0.152 0.160 
DGleftCright P135 1.161 1.165 
DGrightCleft 1.125 1.108 1.075 
DGG +117 1.137 1.118 
DCC 1.138 1.117 1.104 
  
  
  
  
  
where, D —distance, G- GPS antenna, C = camera 
As shown in Table 3, the effect of radial lens distortion is 
significant. Inclusion of the Kl parameter results in an 
improvement of between 3096 and 5096 over not using the 
parameter. That means that the geometric distortion of current 
CCD camera is an important factor what must be considered. 
About the effect of higher order terms of lens distortion will be 
investigated in the coming experiment. In this experiment the 
accuracy of control point coordinates can reach only 5mm and 
the accuracy of image coordinates measurement is about half 
pixel. Due to the limitation of the camera resolution and the 
blurredness effect, even though imaging in the evening to 
enhance the background/foreground contrast, the accuracy 
improvement of image coordinate measurement is still very 
insignificant. 
Table3 The accuracy assessment of calibration (m) 
  
  
  
  
  
  
DET Self-calibration 
# kl ki1+k2 | kl+k2+pl+p2 
N | 0.014 | 0.014 0.007 | 0.008 0.008 
E 10.072 0.063 0.054 | 0.053 0.055 
H- [0.012 0.010 0.009 | 0.007 0.008 
  
  
  
  
  
  
  
The RMSE is computed based on the check points which are 
not included in the calibration process. 
4.3 Multi-image Simultaneously Processing 
Since the imaging baseline length is limited by the mounting 
bar, it is hard to improve the accuracy of spatial intersection 
depending on only one stereopair. The same three stereopairs 
are then processed simultaneously. The resulting relationships 
between GPS and Camera are shown in Table 4. 
Table 4 Relationships from multistation adjustment (m) 
  
  
  
  
  
  
  
  
arameter station | station2 station3 
DGleftCleft 0.1215 0.1222 0.1205 
DGrightCright 0.1720 0.1588 0.1784 
DGleftCright 1.1338 1.1689 1.1560 
DGrightCleft 1.1269 1.1088 1.0874 
DGG 1.1169 1.1365 1.1182 
DCC 1.1272 1.1253 1.1072 
  
  
  
  
Comparing these results with those of Table 2, the RMSE 
(computed from the difference of the offset and its average 
value from three stereopairs) of the latter is 0.0108m more less 
than that of 0.0175m. Also the accuracy of spatial intersection is 
improved evidently as shown by the comparison of Table 3 and 
Table 5. For example the relative accuracy of position (N.H) 
comparing with the size of object is improved from about 1/310 
to 1/700, and the relative accuracy of depth is also improved 
from 1/180 to 1/1100. 
Table5 The accuracy of multistation adjustment (m) 
  
  
  
  
  
RMSE Case I Case II 
N 0.0092 0.0049 
E 0.0250 0.0091 
H 0.0034 0.0029 
  
  
  
  
where, in the Case I the three camera stations are in a horizontal 
plane, and in another case they are in a vertical plane. 
4.4 The Effect of Imaging Configuration on the Accuracy 
Because the network configuration is also very important to the 
accuracy of intersection, in order to achieve higher mapping 
precision by multi-station imaging, careful attention must be 
paid to the design of the photogrammetric network (Fraser, 
1996). In order to analyze the effect of configuration on the 
photogrammetric accuracy, there were two different station 
configurations designed. The first one including three 
stereopairs is illustrated in Figure 5, in which he height 
differences of these three stations were very small, i.e. they 
were set almost in a horizontal plane; another group of 
convergent images were taken as shown in Figure 6, in this case, 
the stations were set in a vertical plane. Since taking more 
robust imaging configuration, the accuracy is increased by 
about 4096 to 6096 as shown in Table 5. In such case of 
configuration, the relative depth accuracy can reach 1:1100. 
From Table 6, it is also easy to see that the baseline/distance 
ratio is a very important factor of configuration to the 
photogrammetric data quality. To increase the baseline will 
cause to decrease the overlap, so in our experiment the largest 
convergent angle can be only about 42 degrees. On the other 
hand, the distance from object to camera is also the key factor 
on mapping accuracy. However because of the limited 
dimension of the test site on the building top, such experimental 
analysis with respect to the imaging distance has not been done. 
256 
Fi 
  
  
  
  
Using the 
baseline/di 
single-stere 
But if t 
adjustment 
increased t 
geo-referer 
practice ai 
obtained c 
bundle ad 
achieved. | 
constructic 
This rese: 
University 
Lawrence 
processing