ABSTRACT
A hidden Markov model (HMM) is a special class of Markov models where the system under study is modeled as a Markov Chain, but the actual states of the system and other parameters are not observable; i.e., they are hidden, and thus the name of these models. A system of sensors tracks people as they move through a museum, which consists of five rooms, four exhibition rooms and a foyer. There are various types of Markov models depending on whether the underlying time is discrete or continuous, and whether the set of the states is discrete or continuous. The forward–backward algorithm, the Viterbi algorithm, and the Baum–Welch estimation procedure described in the previous sections have been extended to the case of continuous observation densities. An underlying assumption in the HMMs examined above, is that the successive observations are not correlated. There are real-life cases where the successive observations are correlated, such as the air temperature measured every minute.
