75 
wind 
1 anemometer 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
   
  
eS. video signal 
  
  
  
  
  
  
  
  
  
light sheet É 
camera in 
underwater box 
  
Figure 1: Scheme of the optical instruments used at linear flume in Delft: A camera in an underwater box is looking 
perpendicular on a light sheet, illuminating seeding particles. 
  
ter 
to 
les 
by 
ive 
Ver Figure 2: Seeding particles at a wind speed of 4.2 m/s (left) and 6.4m/s (right) beneath the water surface in a light 
S sheet illumination. Due to the exposure time of the camera particles in motion are visualized as streaks. 
or 
3 DETERMINATION OF THE 2-D FLOW VECTOR FIELD 
Several steps of image processing are required for the extraction of the flow field from the image sequences. 
g g ge seq 
After the digitization of each picture is parted into its two image fields. A specially developed segmenta- 
tion technique for identifying individual particles from the background follows. Each object is labeled and 
q g 
1d. finally the correspondence problem of identifying the same object in the next image frame is solved by calcu- 
y p g 8 
lating its streak overlap. Repeating this algorithm will track segmented particles through the image sequence. 
1c 
4 . 
H 3.1 Preprocessing 
300 An image taken with a standard CCD camera at a frequency of 30 Hz (NTSC-norm) actually consists of 
mH. two consecutive fields of only half vertical resolution. Therefore it is required to split the original gray value 
image g(x,y) into its two fields; one field g;(x.y) being the odd and the second field g2(x,y) being the even 
rows of g(x,y). 
EX 3.2 Segmentation 
see 
ed The histogram (Fig. 3) of a streak image shows two distinct maxima, at the low gray values being faint 
740 particle streaks and the background and at high gray values being reflections at the water surface and bright 
ne, particles. Therefore the intensity of the streaks ranges from the very low to the very high gray value. Simple 
00) pixel based segmentation techniques cannot be chosen as the streak images do not show a true bimodal 
distribution in the histogram. A region growing algorithm was developed for the discrimination of individual 
particles from the background. Regions with similar features are to be identified and merged together to a 
NJ 
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop "From Pixels to Sequences”, Zurich, March 22-24 1995