Modeling Method of Human Actions with HSMM and If-Then-Rules Considering Readability

Kohjiro Hashimoto, Kae Doki, Shinji Doki, and Tohya Ohtsuka


Human action modeling, Hidden semi markov model, Extraction of frequenct time series pattern, Time series signal processing


Systems that assist human operation have been desired. In order to realize such systems, the system must have a certain human action model. Therefore, we propose a modeling method of human actions. In this method, a human action model is generated based on the stored data obtained by observing human actions. In this way, complicated human actions are able to be modeled without the previous knowledge about a task. However, the stored data obtained by long-term monitoring of a person's actions includes the data unrelated to a task. Therefore, the generated model is hard to analyze for designer. According to this reason, it is focused on readability of human action model for designer. In this method, the causalities between human action and the situation around a person are evaluated and extracted from the stored data. The extracted causality is expressed by If-Then-Rule explicitly. Moreover, a time series information of the human action and the situation is modeled with Hidden Semi Markov Model(HSMM). In order to express the time series information by directly the time series data, HSMM has high readability. Therefore, a human action model with high readability is generated based on this method.

Important Links:

Go Back