Motion Database Retrieval with Application to Gesture Recognition in a Virtual Reality Dance Training System - Multimedia Database Retrieval: Technology and Applications - page 311

Table 11.2 Definition of the

P 1 P 2 P 3 P 4 P 5 P 6
P 1 − G 12 G 13 G 14 G 15 G 16
P 2 G 21 − G 23 G 24 G 25 G 26
G 3 G 31 G 32 − G 34 G 35 G 36
P 4 G 41 G 42 G 43 − G 45 G 46
G 5 G 51 G 52 G 53 G 54 − G 56
P 6 G 61 G 62 G 63 G 64 G 65 −
Note : P i is the i -th posture. G ij is the gesture
created by changing from the i -th posture to
the j -th posture. G ji is the reversal of gesture

30 gestures

1if x

true
0 otherwise

=

ʴ ( x )= (11.13)

C since the
SSOM has C nodes. The PT feature vector is formed by arranging the elements of
this matrix, p h ,

BasedonEq.( 11.12 ), the dimension of the transition matrix is C ×

C 2 , into a 1-D template.
In addition, we can also obtain the posture transition sparse codes (PTSC) that
are analogous to the sparse codes of postures, only differing in that they represent
the existence of transitions rather than the frequency of transitions.

h = 1 ,...,

11.5.4 Performance Comparison

The experiments were conducted on an isolated gesture database to assess the per-
formance of the SSOM posture space representations, gesture template definitions
and matching criteria. Based on the six postures discussed previously in Table 11.1 ,
a new set of gestures is defined, containing a total of 30 gestures and 600 instances.
Table 11.2 shows a matrix describing the definitions of all gestures. In the table,
given the six postures P 1 - P 6 , a gesture G ij is formed as an isolated gesture moving
from the i -th position to j -th position (i.e., moving from posture P i to posture P j ).
This definition forms the gesture set, Set I, in the upper triangle of the matrix,
containing G 12 G 56 , which has a total of 15 gestures.

By contrast, the gesture G ji is the reversal of the gesture G ij . The reversal gestures
forms the gesture Set II, which contains gestures in the lower triangle of the matrix.
The total number of gestures from Set I and Set II is 30 gestures.
The non-reversal gestures in Set I were firstly used. Two datasets were con-
structed: Teacher dataset and Student dataset. The database includes 15 isolated
gestures (i.e., each gesture was recorded independently of any sequence of other
movement/gestures). The structure of this dataset is summarized in Table 11.3 .From
the full set of Teacher gestures and Student gestures, 50 % were randomly selected
and used to form gesture templates, while all 100 % were classified against these
templates.

,..., G 16 ; G 23 ,..., G 26 ; G 34 ,...,