|BaseHiddenMarkovClassifierTModel, TDistribution, TObservationThreshold Property|
For gesture spotting, Lee and Kim introduced a threshold model which is composed of parts of the models in a hidden Markov sequence classifier.
The threshold model acts as a baseline for decision rejection. If none of the classifiers is able to produce a higher likelihood than the threshold model, the decision is rejected.
In the original Lee and Kim publication, the threshold model is constructed by creating a fully connected ergodic model by removing all outgoing transitions of states in all gesture models and fully connecting those states.