xpect 
h the 
r less 
with 
ately 
2000 
  
Martin Smith 
Curve | (model data) = 1768 points (average 3.1m between scaled points) 
Curve 2 (ground data collected on foot) = 8184 points (average 0.8m between points) 
Number of matched points = 444 points 
RMSE plan = 0.21m 
RMSE height = 0.19m 
Setting this model up using traditional ground control for absolute orientation (9 points) gave RMSE values in plan = 
0.62m and in height = 0.31m. So the curve matching technique has provided significantly better results. This is 
probably due to the very high density and therefore a large number of matched points. The matched points (linear 
features) also have a very good distribution. 
5.1.3 Llangollen, North Wales photograph results. Again 1:10 000 scale photography was used, but this time a 
vehicle collected GPS data around a road network in Llangollen that consists of very rugged terrain (see figure 6). On 
occasions the vehicle reached speeds in excess of 40mph. To add practical reality, the photogrammetrist was not 
involved in the field data collection and attempted to observe the same route in the stereo model on the SD2000 from a 
brief written description. 
  
  
[roses 
Y Be 
1 
— 2004 N 
5 
T 150 J : MA. 
™ 
180 N ( 
50.1 : Y 
mee — 2 À Le EST 
add wt. Dem 
: Sie mr Lf — 25 4 
Eni rd ed, a 
x 10° 343 To Lee 328 : 
ENT 2322 10° 
Northings Eastings 
Figure 6. Perspective view of the GPS data collected at Llangollen 
(values in metres, note the exaggerated Z scale) 
  
  
  
Curve 1 (model data) = 435 points (average 12.9m between scaled points 
Curve 2 (ground data) = 3527 points (average 2.8m between points) 
Number of matched points = 355 points 
RMSE plan = 9.3m 
RMSE height = 4.5m 
The small number of model points and the large changes in the relief has resulted in a match/transformation worse than 
expected. Using traditional ground control in this stereo model the absolute data collection and thus the orientation is 
not particularly good with RMSE values X=2.6m, Y=2.6m and Z=0.54m. A further effect which has contributed to the 
larger than expected RMSE values is the difficulty for the two curves to follow exactly the same track for example, the 
vehicle was obviously limited in its route line by the road conditions. Taking this into account the result is perhaps not 
SO surprising. 
5.2 Exterior Orientation 
The curve matching technique is ideally suited to matching curves of like dimensions; 2D to 2D and 3D to 3D curves. 
In the process of exterior orientation we are relating 2D image to a 3D real world coordinate system. The relationship 
between the 2D image and 3D world is given directly and efficiently by the collinearity equations. An example of 
taking 2D image coordinates (Z value held constant at zero) and fitting to a 3D curve using the technique developed 
produces a typical result shown in figure 7. This example comes from part of the data set for the Nottingham University 
Campus photography. The following results were obtained: 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 855