Authors: Sascha Bosse, Claudia Krull, Graham Horton
Gesture recognition is an important subtask of systems implementing human-machine-interaction. Hidden Markov Models achieve good results for gesture recognition in real-time supporting a low error rate. However, the distinction of gestures with different execution speeds is difficult. Hidden non-Markovian Models provide an approach to model time dependent state transitions to eliminate these problems. In this paper, a basic non-Markovian model structure for gesture recognition is developed. The experiments show that Hidden non-Markovian Models are not only applicable in the field of gesture recognition, but that they can also distinguish gestures with different execution speeds.