Substructure and boundary modeling for continuous action recognition

摘要

Embodiments of the present invention include systems and methods for improved state space modeling (SSM) comprising two added layers to model the substructure transition dynamics and action duration distribution. In embodiments, the first layer represents a substructure transition model that encodes the sparse and global temporal transition probability. In embodiments, the second layer models the action boundary characteristics by injecting discriminative information into a logistic duration model such that transition boundaries between successive actions can be located more accurately; thus, the second layer exploits discriminative information to discover action boundaries adaptively.

主权项

1. A computer-implemented method for estimating a sequence of action labels given an observation sequence, the method comprising:
given a linear dynamic system state space model that is capable of being represented as a unified probabilistic graphical model comprising duration-of-actions layer (D) consisting of a series of time duration values, an action label layer (S), a first hidden state layer comprising a latent state layer (X), a second hidden state layer comprising a substructure transition layer (Z), and an observation layer (Y): maximizing a joint posterior of action label at each of a set of time instances given a set of observations; approximating the joint posterior using sampling, wherein a number of samples are drawn to approximate the posterior; and calculating the posterior by weighted combination of the samples; wherein determination of a new action label in the action label layer (S) and new substructure transition in the substructure transition layer (Z) are triggered by the ending of a time period defined by a time duration value in the duration-of-action layer (D).