low. Due to a suboptimal quality of the calibration the tolerance
to the epipolar line has to be increased which leads to a higher
number of ambiguities when the previous method is applied.
With the new implementation the number of links could be
increased by more than 40 % (Fig. 8).
20"
TBP en
er
figure 6: Particle trajectory reconstructed with previous
jementation (left), same trajectory reconstructed with new
spatio-temporal matching algorithm (right)
— ditionally calculated particle positions are marked in
y (Fig. 6, right) Of course not every trajectory
suction could be improved in that impressive way, but
dosing a gap of one or two time steps is a substantial
xement for the hydrodynamic analysis.
7. RESULTS
tint
]
lio
iult of the 3D PTV processing is a velocity field with a
d spatial resolution and a low number of mismatched
iles. A visualization of results is shown in Figure 7 (only a
tion of trajectories is displayed for the sake of better
iit).
10-temporal matching:
) available data used by the
1ethod, (C) result of the new
ased tracking
Figure 7: Velocity field (visualization of results)
ist the new algorithm and to compare the results achieved
ithe previous implementation three different data sets were
itable, the loss of quality Sessed, The data sets ‘Copper sulfate’ and ‘Triocular’ were
compared to an interruption yired during the observation of real experiments, while data
olation is only used for Simulated vortex’ was completely synthetically generated.
of the particle positions Oe data sets were processed with the previous and the
space observations. ^ method, The processing with the new algorithm was
mple from an experiMilnyeq in forward direction and also bidirectional, which
king method in comparison ll further improvement.
he previous implementation
ıld only be reconstruct
the new spatio-tempordl y Data set “Copper sulfate’
rticle without gap over {he
ttubulent flow in an aqueous copper sulfate test fluid
"Wen two electrodes was recorded with four cameras (Lüthi
4l, 2001). In this experiment the seeding density was rather
600 ,
500 |
400 |
300 |
200 |
Number of particles
100 |
0
Previous New (forward only) New (bidirectional)
ElUnlinked 95 67 64
El Linked 313 411 446
Figure 8: Results of the tracking methods (copper sulfate)
7.2 Data set ‘Triocular’
A three camera setup was used to record the flow field in this
experiment. Image quality and calibration accuracy were quite
high, nevertheless the epipolar intersection led to a high number
of ambiguities which may be solved by a four camera setup.
Making use of the spatio-temporal context these ambiguities
could be solved even when only three cameras are used for the
image acquisition. The number of links was increased by almost
40% while at the same time the number of unlinked particles
was reduced (Fig. 9).
800 -
700 |
600 |
500 |
400 |
300 |
200 |
100 |
Ü lon
Number of particles
1 New (forward only) New (bidirectional)
69
Previous
BE Unlinked 115 72
El Linked 501 654 699
Figure 9: Results of the tracking methods (triocular)
7.3 Data set ‘Simulated vortex’
The algorithm was also tested on the basis of a synthetically
generated data set simulating a vortex flow seeded with 1500
particles. Image sequences were created assuming a typical
PTV hardware setup with four convergent cameras.
Due to error-free orientation parameters and lenses without
distortion the epipolar intersection method established already a
high number of consistent correspondences. However, applying
the new method a further increase by almost 5% was achieved.
Out of the 1500 particles only about 8% remained unlinked or
undetected (Fig. 10).
—605—
