Proposed Algorithm
These concepts already introduced are used to propose a new architecture,
which is capable to classify from multiple channels having potentially noisy signals
and to take the failure of one or more channels into account. The architecture,
illustrated in Figure 1, treats every available input channel separately and classifies
on a single feature vector. The output of each classifier is monitored and rejected in
case the confidence fall below a certain threshold. Since the remaining results are
often still relatively weak and in order to make a decision for the time segments of
the rejected samples, a temporal integration over a set of individual frames is
conducted. Due to the rejection and