0.10 
0.08 + Hi 
0.06 + 
0.04 1 
0.02 +: 
0.00 ! 
780 820 860 900 
STD (m) 
  
  
  
  
PHOTO NUM BERS 
     
  
€ east i north à height 
  
  
  
Figure 3: Predicted Positioning Accuracy of Reference 
Trajectory 
  
  
0.008 
0.006 + - - - - - - 
0.004 + 
STD (deg) 
0.002 
  
  
  
0.000 
780 820 
  
PHOTO NUMBER 
A roll @ pitch e azimuth 
Figure 4: Predicted Orientation Accuracy of Reference 
Trajectory 
  
  
  
Due to the large overlaps the photogrammetric block is of a high 
redundancy. The accuracy of the perspective centres of the 
photographs is estimated by traditional block adjustment using all 
ground control points. The position accuracy of the perspective 
centres is about 3 cm (STD) in horizontal and 2 cm (STD) in vertical 
direction (Figure 3). The standard deviations (STD) of the orientation 
angles are depicted in Figure 4. Their mean values are about 
2 milli-degree (mdeg) in roll and pitch and 1 mdeg in azimuth. 
Hence, the parameters of exterior orientation are determined with an 
accuracy which is two to four times better than that expected from 
the tested attitude/positioning system. They can therefore be used as 
a reference. 
4. ACCURACY OF GPS/INS EXTERIOR ORIENTATION 
The GPS/INS data has been processed according to the methodology 
discussed in Section 1 by using KINGSPAD (KINematic Geodetic 
System for Position and Attitude Determination), an integration 
software package developed by The University of Calgary. This 
program has been recently extended to employ direct integer 
ambiguity search (DIAS) in either static or kinematic mode (Wei and 
Schwarz 1995). 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
    
Problematic GPS data, containing either a number of cycle slips, or 
having more than one loss of lock or poor satellite geometry, was 
processed first in wide-lane mode with frequent kinematic OTF 
ambiguity search. By using the two frequencies for widelaning, a 
noisy but unbiased flight trajectory could be established, which then 
served as a reference for final single frequency processing. During 
periods of time when only the same four satellites were visible, the 
INS data was the only source to assist in cycle slip detection. The 
64 Hz program navigation output was linearly interpolated to obtain 
position and attitude for the camera exposure time. 
After correcting for the spatial offsets between sensors, a comparison 
between parameters of exterior orientation derived from GPS/INS 
and from the photogrammetric bundle adjustment could be made. 
The position differences reflected in Figure 5 have RMS values of 
15 cm horizontally and 20 cm vertically. Since the separation 
between the rover and master station receivers was within 10 km, 
these rather large differences are most likely due to the small number 
of observable satellites and the poor satellite configuration. In order 
to achieve reasonable geometry, low elevation satellites had to be 
included into the processing. Even with an elevation mask as low as 
10 degrees, only 4 to 5 satellites were simultaneously tracked by both 
receivers. 
  
  
  
  
  
  
     
  
  
  
  
  
  
  
  
  
  
  
  
1 
E 
z 
9 
E 
o 
o 
a 
-1 
20 30 40 50 60 
TIME (min) 
+ east m north à, height 
Figure 5: Variation in GPS/INS-Camera Position 
0.15 
—- 0.10 + 
o 
$ 0.05 + 
a 0.00 À b 
E -0.05 + 
E -010- 
< 
-0.15 
24 30° 31 41 46 52 58 
TIME (min) 
| roll = = = = pitch azimuth | 
  
  
  
  
Figure 6: Variation in GPS/INS-Camera Orientation 
The differences between the orientation parameters are depicted in 
Figure 6 by means of the three Euler angles roll, pitch and azimuth. 
The time scale of this figure is not uniform due to the fact that 
comparisons can only be made during the short time periods when 
128 
  
  
  
  
  
  
  
  
  
  
  
  
   
  
   
  
  
   
   
  
  
   
   
   
   
  
  
  
  
   
  
   
  
  
   
   
  
   
  
  
  
   
  
   
    
flying over 
distributed wi 
mission. Th 
INS-derived 
frequent GP 
effective in 
explanation, 
roll. Both are 
in pitch and s 
to several rea 
synchroniza 
noise due 
photogramm 
to the strong 
a mixture of : 
attitude erroi 
the unexpec 
Target point 
unknowns 
determined 
well as the 
calibration ar 
GPS/INS in 
georeferenci 
means of ] 
Practically, 
parameters 
coordinates 
control and | 
GPS/INS. T 
points are de 
deviation of 
the points a 
Since the ca 
parameters 
directly prop 
the errors sk 
resulting p 
might indi 
photogram 
deviations, 
GCP ERROR (m) 
  
  
Figure