International Archives of the Photc 
  
  
  
  
Figure 12. Barrage structure measured from the panoramic 
stereo model in Figure 11. 
3. RESULTS AND DISCUSSION 
In order to verify the procedure, we compared geometries of 
some structures, which were measured both with tacheometer 
and from photogrammetric documents. The structures were one 
barrage system (B1) and two terrace Systems (T1 and T2). The 
measurement geometry is presented in Figure 13. The barrage 
system and its measured points are shown in Figure 12. 
As a kind of rule of thumb one can expect precisions of stereo 
photogrammetric measurements using following formula: 
  
  
a = 2 . 3X 
c : (1) 
zZ 
dY = . dy 
e ,and (2) 
E m 
ua =". 
eB 3) 
in which X and Y are the coordinates parallel to the image 
plane, Z is the coordinate in viewing direction (i.e. the 
distance), C is camera constant, B is the effective base of 
photography. 
We consider here roughly that the precision dx or dy of any 
measured image coordinate is one pixel and the precision of the 
parallax measurement dpx is 1.4 pixels. The measurement 
geometry in our experiment is presented in Table 1 and Figure 
13. and the expected precisions in Table 2. In case of our 
experiment, the camera constant C is 1400 pixels, the base B is 
7.5 m and the distances to measured structures varied between 
45 — 100 m. 
The results indicate that panoramic images can be applied for 
archaeological survey. With regard to measured locations of 
structures, the external accuracy meets the expectations: 
Barrage Bl in Tables 2 and 3, and all structures in Figure 13. 
The measured bias in case of barrage Bl, i.e. the mean of 
coordinate differences between both measurements are within 
the range of expected precisions. 
With regard to measured points, the internal precision meets the 
expectations as well: Barrage Bl in Tables 2 and 4, and in 
Figure 14. Both measured shapes of Barrage BI appear similar, 
whereas in case of Terrace T1, the shapes become scattered 
when measured from images. However, the repeatability 1s 
good. This indicates that the interpretation on images becomes 
»grammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part 
BS. Istanbul 2004 
vague or uncertain compared to the interpretation in sifu with 
tacheometer. 
Table 1. The measurement geometry. 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
c B dx dy dpx 
[pixel] [m] [pixel] [pixel] [pixel] 
1400 7.5 ] 1 1.4 
Table 2. Expected precisions in the experiment. 
Z dX dY dZ 
[m] [m] [m] [m] 
Barrage Bl 45 0.03 0.03 0.27 
Terrace T1 60 0.04 0.04 0.49 
Terrace T2 100 0.07 0.07 1.35 
1 / ss 
sat Jj > 
vef. ol 
p- À 
if 
  
  
  
Figure 13. Measurement geometry of presented 
photogrammetric survey of terrace and barrage structures. For 
comparison, the structures are measured with tacheometer. The 
base of the stereo photography is 7.5 m and the distance here to 
the furthest structure is 100 m. 
Table 3. The bias in Barrage Bl. 
  
  
  
N (-X) Z (-Y) E (-Z) 
[m] [m] [m] 
Bias -0.01 0.11 -0.17 
  
  
  
  
  
When imaging distances increase, the uncertainties in measured 
distances increase. According to the Formula 3 the effect is 
squared with respect to the distance (Z). This effect becomes 
clear in case of Terrace T2 (Fig. 15) where the shapes become 
highly scattered when measured from images. 
However, with respect to repeated image measurements 1 and 2, 
the range of scattering is higher than the expected 1.7 m in 
Table 2. This increased scattering can be explained by higher 
uncertainty in finding exactly corresponding points on stereo 
images when the structure is lineated parallel to the image base. 
Instead of point wise measuring, the structures should be 
interpreted on images as linear features, which would lead to 
less scattered shapes. 
In case a DEM is available, these linear features can be 
measured on single images and the 3-D determination would 
proceed by intersection with the DEM surface. This alternative 
method is called mono plotting (Viite). However, mono plotting 
is inaccurate, since it assumes that the structures lie flat on 
terrain. 
The proper way would be to measure the structures as 
corresponding linear features on at least two images and then 
determine the corresponding 3-D structure by intersection in 
space.