82 
To obtain sequential information from video tape requires very fast storage of data. A means of obtaining such information is to 
encode the digitised images in JPEG Movie format and then sequentially extract them for processing. A frame-grabber with these 
capabilities was used to store sequences from past tests. A JPEG movie viewer was created for use in the X Windows environment 
to allow for sequential playback of individual frames from the JPEG movie sequence. Each image was then processed in the 
following manner. 
(a) The target image locations were detected using a intensity feature based method. The poor quality of the target images 
and the variable background illumination meant that reliable location of the targets. was difficult using conventional 
methods. Hence, a simple but effective algorithm was written in which a 5x5 mask was moved around the image and the 
characteristics of an intensity peak caused by a target were found. A target was located by taking the intensity of the point 
in the image at the centre of the mask and searching for pixels in the eight locations surrounding this point. If the average 
of their differences was around seven or greater grey values then this point was considered a candidate target. The sixteen 
pixels surrounding the previously used eight were then checked to establish whether their average difference from the eight 
was around five or greater. By combining both criteria it was possible to detect the majority of the targets in the video 
images. The advantage of this method is its independence from the background level at a relatively modest computational 
cost. 
(b) After shape analysis and threshold determination the centroid of the target was computed and the locations of the 
targets from the first image passed to the second image where the centroid of the target was again computed and passed to 
the next image, and so on. In this way targets were tracked from image to image with minimal errors. The use of this 
procedure meant that no new targets could be added in subsequent images and there is a possibility of targets deteriorating 
and causing misidentification. However, tracking of targets with few errors of any consequence proved possible. 
The current method of data collection suffers from two important limitations: poor image quality and the fact that analysis of images 
has to be made after the end of the experiment. Hence, further investigations were required to improve the operation of the system 
which are described in the next section. 
3. IMPROVEMENT OF THE ORIGINAL SYSTEM. 
3.1 Calibration 
Any transformation from image measurements to 2-D object co-ordinates within the centrifuge is subject to two major sources of 
.error. First, the sample box and the camera can move as the centrifuge is spun up to speed and second, the camera has an optical 
system which, by virtue of its short focal length, was subject to gross barrel distortion. The original video image did not contain any 
control points. Consequently no physically appropriate model for positioning could be directly applied to the data without more 
information. Calibration, based for example on the plumb-line method, was considered but no suitable colour frame grabber was 
available to obtain images directly. The only dimensional information available derived from the fact that the targets were positioned 
in the soil using a template. Comparisons between the target template locations and image data obtained once the centrifuge was at 
test speed, but before the test itself had started, could be used to generate a deterministic mathematical model. The statistically 
significant parameters of a third order polynomial were used to model the data by the method of least squares. The polynomial could 
then be applied as a rudimentary system calibration to correct subsequent target image measurements. Whilst giving no better than 
1mm. standard deviations for object space positions, the procedure could at least be used for extracting photogrammetric data from 
tapes obtained before the use of photogrammetry had been considered by the geotechnical engineer. These computations have 
nevertheless allowed useful results (Figures 5 & 6) to be obtained from the archive tapes. Once the imaging system has been 
redesigned, a full component calibration will be carried out, based on physical and photogrammetric principles, and then used as an 
integral part of the geotechnical experiment. 
  
  
  
  
  
  
  
  
a N El HE T8 
2]: -- 10 DT = : TIN of 
> - 7 - = NUN NN NS. -8 pos - - “+ -8 
= mr bl] Ng NN LA 3 1 2 d NUUS : 
= my WEN iG > - oo S N 4 — 
Hille wim wr =~ 4: H82 i "> + > > 35 -8 9 
O ee ee ug gg ‘ e Ö e ot sq 
«il. -- --—— «L.X «el ::---:-- | X 
d e ei ea A iu unt dr i 2e d -— eu a ue at. o & 
Bl AR t & ftro 1 BE 
a Te = = - -— 7 * F a = = 
mE wwe ‘ 7 
Hz SE r2 m : [2 
gi---—t1:í:' L a i 
El: ts z 
Blasio F8 = Ei 
Qi: i a I 
" T Do ed VE TT T ° 
9; 8 3 3 3 ^? 2. 8 8 8 *. 
(siexid) A (sjox1d) A 
Figure 5. Before correction. Figure 6. After correction. 
IAPRS, Vol 30, Part 5W1, ISPRS Intercommission Workshop "From Pixels to Sequences", Zurich, March 22-24 1995 
  
lig 
the 
us 
pe 
ins 
rer 
the 
ori 
als 
To 
ligl 
foc 
est: 
spli 
mal 
is n 
14