Human motion analysis using the rhythm-an estimate method of dance motion with multivariate autoregressive model

In our studies, we have focused on the traditional dance of Japan recorded in an image database. In this study, we aim to reproduce the traditional dance motion recorded in an image database. We propose a method of estimating dance motion with the multivariate autoregressive model. The multi-articular motion is a multivariate feedback system. We analyze the motion of three-variables (shoulder,elbow and wrist). As a result, we confirm that the wrist motion is controlled for the shoulder motion in the low frequency region. The elbow motion has been separated without affecting the other joint. Using the multivariate autoregressive model, we estimate the motion with the effect between joints and get good results for estimation of dance motion.