STEREOSCOPIC 3D-IMAGE SEQUENCE ANALYSIS OF SEA SURFACES 
F. Santel, W. Linder, C. Heipke 
Institute of Photogrammetry and Geolnformation, University of Hannover, 
Nienburger Straße 1, 30167 Hannover, Germany 
(santel, linder, heipke)@ipi.uni-hannover.de 
Inter-Commission V/III 
KEY WORDS: Matching, Sequences, Sea, Surface, Reconstruction, Video, Stereoscopic, Three-dimensional 
ABSTRACT: 
Numerical modelling of highly complex processes in the surf and swash zone is an important task in coastal zone management. As 
input and reference data for these numerical models three-dimensional information about the water surface is required. In this paper 
a method for reconstructing a dynamic digital surface model of a surf zone based on stereoscopic image sequences is presented. The 
surface model is obtained by digital image matching using a variation of the vertical line locus method. The processing principles for 
stereoscopic image sequence analysis and the results are described. The image matching is checked by manual stereo analysis and 
gauge data. The research area is a groyne field on a North Sea island in Germany. 
1. INTRODUCTION 
In coastal zone management the optimisation of constructions 
like dykes or groynes is of high interest. The design of their 
shape and surface properties requires detailed information about 
the waves attacking them. In this context monitoring and 
prediction of the sea state in the surf zone is very important. 
The processes in the surf zone, like wave breaking, wave runup 
and wave overtopping can be described by numerical modelling 
(e.g. Strybny, Zielke, 2000). The geometric shape of the water 
surface is an important element for the numerical models. 
However, currently only point-wise gauge and buoy 
measurements are available to control such models. 
In principle the water surface model can be provided with the 
required temporal and spatial resolution for the calibration and 
validation of the numerical model using digital photogrammetry 
(Strybny et al, 2001). Digital image matching was already 
employed successfully for the determination of wave 
parameters from stereo images in the past (e.g. Redweik, 1993). 
Further examples for the determination of sea surfaces using 
stereo images are given in Holland et al. (1997), Taguchi, Tsuru 
(1998) and Yamazaki et al. (1998). 
The goal of our work is the three-dimensional and quasi- 
continuous determination of the water surface in the surf zone 
using an automatic photogrammetric approach. The approach 
processes stereoscopic video image sequences by image 
matching. 
The analysis of image sequences is a challenging problem and 
has been an important research topic in the areas of 
photogrammetry and computer vision for some time. Horn 
(1986) for example used optical flow to determine the motion 
of a camera from an image sequence. An algorithm obtaining a 
three-dimensional model from image sequences is presented by 
Pollefeys et al. (2000). The system is able to extract 
automatically a textured three-dimensional surface from an 
image sequence without prior knowledge about the scene or the 
camera. In our case image sequence analysis is used for the 
surface determination of a dynamic process, i.e. the tracking of 
a moving surface with static cameras. 
2. IMAGE MATCHING 
The computation of a digital surface model from images 
requires the interior and the exterior orientation of the images as 
well as homologous points. Assuming the orientation to be 
given, the identification and the image coordinate measurement 
of homologous points in two or more overlapping images via 
image matching over time remains the major task to be solved. 
2.1 Point-wise Correlation 
The three-dimensional determination of the water surface is 
accomplished by digital image matching using stereoscopic 
images as implemented in the software package LISA (Linder, 
2003). By successive point-wise matching based on cross 
correlation over the model area through a sophisticated region 
growing algorithm starting from given seed points, a three- 
dimensional point cloud is generated, subsequently a digital 
surface model (DSM) is obtained by interpolation. 
- EX 1 7 s 
RT. | 
N \ ' | 
\ 
À A. 
L 
Figure 1. Point-wise matching algorithm 
   
    
  
   
    
  
    
    
  
  
  
    
    
  
    
   
   
   
   
      
   
  
  
     
   
   
    
    
     
   
    
  
   
    
The 
Api 
p 
Orie 
line 
CC 
poii 
whi 
sev 
spa 
Usi 
pro 
Squ 
pos 
def 
ima 
cas 
of 
Wir 
to | 
in 
abc 
rej 
In : 
for 
nei 
mil 
anc 
rej 
Th 
COI 
(Be 
lin 
2.2 
poi 
un 
In 
thi 
foi 
art