Discrete hidden Markov models with application to isolated user-dependent hand gesture recognition

The development of computers and the theory of doubly stochastic processes, have led to a wide variety of applications of the hidden Markov models (HMMs). Due to their computational efficiency, discrete HMMs are often favoured. HMMs offer a flexible way of presenting events with temporal and dynamical variations. Both of these matters are present in hand gestures, which are of increasing interest in the research of human-computer interaction (HCI) technologies. The exploitation of human-to-human communication modalities has become actual in HCI applications. It is even expected, that the existing HCI techniques become a bottleneck in the effective utilization of the available information flow. In this work it is given mathematically uniform presentation of the theory of discrete hidden Markov models. Especially, three basic problems, scoring, decoding and estimation, are considered. To solve these problems it is presented forward and backward algorithms, Viterbi algorithm, and Baum-Welch algorithms, respectively. The second purpose of this work is to present an application of discrete HMMs to recognize a collection of hand gestures from measured acceleration signals. In pattern recognition terms, it is created an isolated user-dependent recognition system. In the light of recognition results, the effect of several matters to the optimality of the recognizer is analyzed.