Database Reference
In-Depth Information
Fig. 11.6 Online recognition of Teacher dance gestures; left top : posterior traces based on posture
occurrence; bottom left : class prediction trace for posture occurrence; right top : posterior traces
based on posture transitions; bottom right : class prediction trace for posture transitions
Although the errors introduced by feature extraction algorithms or deviations in
user performance do not influence the entire trajectory, the errors do influence the
evaluation of a particular node or transition.
11.7.1
The Self-organizing Map Distortion Measurement
The structure of SSOM has various configurations, including c0, c1, c2a, c2b, c3a,
and c3b; each has a different number of map nodes, from 12 to 642, respectively.
These configurations constitute a map for vector quantization with a distortion,
resulting from topologies and number of nodes [ 342 ]. In the current work, the SSOM
is expanded to have multiple codebooks that are designed according to gesture class
characteristics, instead of only one codebook.
Search WWH ::




Custom Search