Database Reference
In-Depth Information
0.8
0.75
0.7
0.65
0.6
0.55
Whole Sequence
Sample 80%
Sample 60%
Sample 40%
Sample 30%
Sample 25%
Sample 20%
0.5
0.45
0.4
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
Delta percentage of sequence length
Fig. 13.
In most datasets, the similarity stabilizes around a certain point after a
value of
δ
.
similarity stabilizes to a certain value (Figure 13). The determination of
ε
is application dependent. In our experiments we used a value equal to some
percentage (usually 0.3-0.5) of the smallest standard deviation between
the two sequences that were examined at any time, which yielded good and
intuitive results. Nevertheless, when we use the index the value of
ε
has to
be the same for all pairs of sequences.
If we want to provide more accurate results of the similarity between
the two sequences, we can use weighted matching. This will allow us to give
more gravity to the points that match very closely and less gravity to the
points that match marginally within the search area of
. Such matching
functions can be found in Figure 14. If two sequence points match for very
small
ε
ε
, then we increase the
LCSS
by 1, otherwise we increase by some
amount in the range
r
,where0
≤ r<
1.
6.2.2.
Experiment 1 — Video Tracking Data
These time series represent the X and Y position of a human tracking
feature (e.g. tip of finger). In conjunction with a
“spelling program”
the
user can
“write”
various words [20]. In this experiment we used only the X
coordinates and we kept 3 recordings for 5 different words. The data cor-
respond to the following words:
'athens', 'berlin', 'london', 'boston', 'paris'
.
