Human Action Modeling Method based on the Causality between a Sitution and an Action

Modeling by Considering Change Speed of Time-series Data with HSMM

Abstract

We have proposed a modeling method of human actions based on the causality between a situation and an action. In this method, a human action rule is expressed by an If-the-rule style, assumed that a person changes his current action to the next one according to the situation around him. In the previous method, a human action and a situation in a human action rule is modeled with a Hidden Markov Model(HMM). HMM is one of powerful tools for modeling time series data, but it ignores the change speed of time series data. In addition, time series data on human actions and situations are classified by Continuous Dynamic Programming. This means that two types of criteria should be set for modeling. In order to overcome these problems, we propose a new modeling method of human actions with Hidden Semi-Markov Model(HSMM) in this paper. In the proposed method, both clustering and modeling of time series data are executed with HSMM. The usefulness of the proposed method is discussed through some modeling results of human actions on operating a radio-controlled vehicle.