f.ch 
f.ch 
's in 
gical 
vn a 
th a 
hich 
ng a 
th a 
and 
ent. 
lem 
it of 
ated 
e in 
'S ls 
sive 
d of 
14). 
ob- 
t of 
lous 
da 
ted 
nth 
om 
1 in 
ded 
  
  
313 
the signal in S-VHS format. The images were recorded in a time lapse mode every 3.2 s. This method allowed 
an observation time of 10 days with a 180 min tape. The chosen recording interval is a good combination of 
observation resolution (the time it takes for snow falling down to the ground) and a “sufficient” recording time 
of the video-cassette. 
In order to certain the snow amount on the branch we measured the position of a branch at a certain time. The 
bending of the branches is a function of the: 
e elasticity modul of the branch under certain conditions (stiffness depending on branch temperature 
(Schmidt and Pomeroy, 1990)) 
e branch diameter as a function of distance from the trunk 
e total amount of snow on the branch 
e snow distribution on the branch 
To measure the position of a branch it is necessary that certain points can be clearly recognized in the image. 
Therefore balls with a diameter of 6 cm were suspended 40 cm below several branches. In the first winter red 
balls were chosen. This color caused difficulties, because around the balls a red comet-like shadow appeared in 
the image which made a clear detection of the balls nearly impossible. 
In the second winter we chose yellow colored balls with 2 W bulbs mounted inside to improve the contrast in 
the image. The tree was illuminated with two floodlights at night. To improve contrast at night the floodlights 
were switched off every 5 min. This gave a well discernable rounded region in the image. To calibrate the image 
several fix points in an absolute coordinate system were mounted in the tree crown. 
Governing factors like air temperature and humidity, global radiation, wind speed were measured in the close 
surrounding of the tree. 
Hardware and software: The analysing equipment consists of a video recorder (Panasonic AG-7355) which 
could be controlled by a computer. Between the video recorder and the computer a time-base corrector (IDEN 
IVT-7P) was inserted to avoid distortion of the video signal which is a known problem when images are digitized 
from a video tape. Hardware consists of a 80486 PC with a Modular Frame Grabber (ITI MFG-3M-V with an ITI 
AM-CLR-VP additional color module) connected to two monitors. A TMS34010 32 bit graphics microprocessor 
is integrated on the frame-grabber board which allows digitizing in real time. Image processing was done with 
the programm OPTIMAS Version 4.01, BioScan Inc. 
2.2 DIGITIZING AND IMAGE ENHANCEMENT 
To analyse time series of images from a video tape it is neccessary to fix certain frames in a sequence. À 
continuous counter on a tape is the so-called time code which is saved either during recording or can be recorded 
later. We recorded the linear time code (LTC) later on the sound track of channel 2 on the tape. So every 
frame had its own number and in our particular case every 10 min a frame could be selected from the computer 
with an accuracy of +1 frame which is the the maximal precision of the LTC. Aspect ratio was measured with 
a rectangular grid and corrected. In order to improve speed of calculation and minimize problems with contrast 
a small region of interest (ROI) for every ball was fixed. After acquiring the image contrast and brightness were 
adjusted and several filter operations were performed. To smooth out noise an average filter was used in order 
to make light intensity more constant over the desired object. An example of three digitized images is shown 
in Fig.la-c. 
2.3 IMAGE ANALYSIS 
To identify the balls in the ROI a threshold consisting of a grey value range for each of the three color channels 
red, green and blue was fixed, which could easily be done interactively by the image analysing program. The 
threshold had to be calculated seperately for several light conditions during day time and at night. Correspond- 
ing to this threshold every ROI could be binarized. To remove undesired pixels an erode filter and afterwards a 
dilate filter was used. Now with an object recognition algorithm the balls could be searched (“search criteria”). 
The algorithm combined searching for a certain area range and a certain circularity of the desired object. The 
center of mass of every ball could be found and its coordinates were written to a file. In words this algorithm 
is expressed as follows: 
“Search every area in the ROI with an area size in the range of 6.8 and 72.8 cm? and a defined circularity of 
the area in a range from 13.5 to 16.5 and write the center of mass of this area to a file”. 
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop “From Pixels to Sequences”, Zurich, March 22-24 1995