anbul 2004 
  
photo 
by manual 
have been 
aging. For 
twice. The 
ts amounts 
ults in the 
eral stereo 
F the entire 
  
41.0 
40 0 
40.6 
E 
Pt 
40.4 
MS 
40.0 
br 
980 
anual 
M) 
] a manual 
the profile 
its a cross- 
urement is 
| is located 
92 m up to 
'ement and 
all wave In 
matching. 
m) due to 
he manual 
rface or on 
m, on the 
matching 
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004 
window and the grey values of the corresponding pixels of the 
right image ridding somewhat different results. In addition, 
some errors appear in the seaward area because of the poor 
texture of the sea surface. The ripples (Y = 943 m up to 951 m) 
as well as the trough of the sea (Y — 958 m) are present in the 
matching result, however the wave at the outer seaward 
boundary (Y = 969 m) can be distinguished, albeit with a wrong 
altitude. 
The standard deviation between the two curves is 9 cm and thus 
in the same range as the analytically determined surface. 
5,2 Multiple Epochs 
Now we turn to the results of the image sequence analysis. 
Three derived water surfaces of the image sequence with a time 
difference of 1 s from epoch to epoch are illustrated in the left 
part of Figure 6. The positions of the wave peaks can be well 
identified and tracked in the surface models. Additionally 
produced orthophotos overlaid to the corresponding surfaces are 
shown in the right part of Figure 6. In both illustrations the 
wave positions coincide. 
  
Figure 6. Sequence of water surfaces with At = 1 s 
left: surface models; right: with overlaid orthophotos 
During the measurement campaign at Norderney Island further 
measurements with conventional instruments, such as current 
meters, gauges and wave rider buoys have been carried out. 
From these point-wise measurements the altitude change of the 
sea surface can be derived. Using these data the developed 
procedure is also checked. 
Figure 7 represents the image matching results at the position of 
one of the gauges and the gauge data over a period of 56 s. The 
56s are equal to a sequence of 450 images acquired with a 
frequency of 8 Hz. The horizontal axis of Figure 7 represents 
the time in MEZ. The vertical axis shows the height in metres. 
The analysed gauge is positioned at the centre of the area under 
investigation. 
As can be seen, the height values determined by image 
matching corresponds to the gauge data in general. However, 
the image matching results show some high-frequency noise 
and problems arise in the areas of poor texture. 
Also, some of the peak values of the two curves differ in the 
range of a few decimetres. It should be noted, that the standard 
deviation of the gauge data is not known in detail and is 
estimated to be in the decimetre range. 
     
    
  
  
  
  
  
   
  
  
  
  
  
  
   
   
   
   
  
   
   
   
  
  
   
    
   
  
  
   
  
    
  
  
  
   
    
    
  
   
   
   
  
   
    
   
   
   
   
   
  
   
   
   
   
   
   
   
  
    
  
  
  
  
  
  
PTS ciem: ue 52 10 
f 46.8 
j I Ai 
; | I! | 
A à || [s 
i i là Fal 
; : s n [| Fd 
| | i P i 4) { | ] i464 
|i {À © A: i] Fo AM 
| EM i | "ri [4 | 3 
i i { fi \} ! , m 
L í i n { | | | | M | | i Un n 
fil Beis of 14 r1 ] "| 131 [Wes 
| À \ | | ‘ | { | pl 1 | £ 
} ld X i | | Lk i / N 
{ \ | V } 1 3 i i y i | 40.0 
fi] | Fad Miri i | 
i \ 1er | Í | ! j \ { 
e ER m E |o des 
i x 4 { | i | j i | i | 
fel \ | | i | | | f | 
be i i ; | N 0 4 
| 3 X / | e \ | 
| ur 2 ¥ | 
BE Um J 9s 
|| — image matching | 1 | 
1 0006 th ct aa itt i 
10:03:56 10:04:05 10:04:13 10:04:22 10:04:31 
time [MEZ] 
Figure 7. Comparison of image matching results and 
gauge data 
In order to further analyse the differences between the matching 
results and the gauge measurements two time steps are 
compared exemplarily. These are time 10:03:58 and time 
10:04:21. 
At time 10:03:58 the gauge is located in a tough of the sea. 
Here a difference of approximately 35 cm between the image 
matching result and the gauge data occur. This difference 
corresponds to 3.3pixel in image space. When further 
investigating this results, we found that the grey value average 
of the matching window amounts 74.1 with a standard deviation 
of 4.8. The mismatch thus occurred due to the poor texture of 
the sea surface. The limits of the algorithm are reached. 
At time 10:04:21 matching is carried out in the swash zone of 
the following breaking wave. The difference between the image 
matching and the gauge data is less than 5 cm. In this case the 
corresponding matching window has an grey value average of 
186.6 with a standard deviation of 15.0. The image matching 
succeeds. 
6. CONCLUSIONS 
The determination of the water surface was carried out by 
digital image matching in object space. An algorithm for image 
sequence analysis was presented. 
Experiments of the developed procedure demonstrate the 
successful derivation of area-wide and dynamic surface models 
of sea surfaces. The analysis of individual stereo models as well 
as of image sequences is feasible. The image matching results 
were spot-checked by manual stereo analysis and by the 
comparison with gauge data and were found to be accurate 
within the expectations. 
Further adjustments of the matching algorithm to the specific 
problems of water surface modelling and the optimisation of the 
parameter choice will be tackled as one of the next steps, 
including tests for reducing the high-frequency noise still 
contained in the matching results. Finally the analysis of the 
entire groyne field will be achieved by combining the two 
overlapping stereo models.