Results (
Vietnamese) 2:
[Copy]Copied!
A framework that naturally incorporates constraints related to the relationship between the states of a system and the uncertainty in the order in which such states occur, is a Hidden Markov Model (HMM) (51-58). In this paper, we adopt HMM to model and infer a time- series of duration-variable atomic activities for each detected worker from a sequence of RGB-D images. In the following, we briefly review the components of HMM before adapting its terminology to our application.
HMMs are dynamic Bayesian networks characterized by three main probability distributions: prior probabilities, state transition probabilities, and emission probabilities [59,60). Assuming thatX = {x,, x2, *T) is a set of T states (hidden or unobserved), and 0 =
{O], o2 or) is a set of Kpossible outcomes for Tstates (also known
as emissions or observations) where the distribution of ot only depends on A',, and xt only depends on xt _ 1, we denote HMM as a 3-tuple X — {A, B, n} where J7 = {rr,} is the vector of initial state probabilities with occurrence rate of each activity, and A = {a¡j} is the state transition prob¬ability matrix that stores the probabilities of transitioning from state x¡
Being translated, please wait..
