摘要 |
PROBLEM TO BE SOLVED: To accurately estimate sympathy interpretation of a dialogue between two persons based on various behavior channels, temporal difference in behavior, and behavioral conformity.SOLUTION: A behavior recognition section 20 uses a learning video which photographs the head of several dialogue persons, as input, to detect behavior of the dialogue persons in the learning video and to output a behavior learning time series which is chronologically labelled. A parameter learning section 40 learns a model parameter containing a timing model which expresses likelihood of sympathy interpretation based on temporal difference in behavior and on behavioral conformity, on the basis of the behavior learning time series and a sympathy interpretation learning time series in which the sympathy interpretation is chronologically labelled using the learning video. The behavior recognition section 20 uses an estimating video which photographs the head of several dialogue persons, as input, to detect the behavior of the dialogue persons in the estimating video and to output a behavior estimating time series which is chronologically labelled. A post-event probability estimation section 50 estimates the sympathy interpretation between the dialogue persons in the estimating video by using the model parameter on the basis of the behavior estimating time series. |