A human motion analysis using the rhythm - a estimate method of dance motion with autoregressive model

Abstract

We aims to reproduce the Japanese traditional dance motion recorded in image database. We propose an estimate method of dance motion using the autoregressive model. In statistical motion estimation with the autoregressive model, the decision of degree is a very important. To decide the suitably degree, we define the "rhythm points" or the time of starting and ending of the motion. We refer to it as the "cycle of every motion" or a number of the frame between three rhythm points. To compare our method with others, we applied two methods: the first method decides the degree from AIC (Akaike's information criterion); and the second method, which we proposed, decides the degree from the cycle of the motion. We introduce the term "decision of suitably degree for a estimate method of dance motion". A good result for an estimate of dance motion was achieved.