Motion retrieval based on an efficient index method for large-scale mocap database

Abstract

In this paper, a novel approach is presented for motion retrieval based on double-reference inde(DRI) in Motion Capture(Mocap) database. Due to high dimensionality of motion's feature, the Isomap nonlinear dimensionality reduction is used. For handling new motion data, Isomap is generalized based on the estimation of underlying eigenfunctions. Then DRI is build based on selecting a small set of representative motion clips in the database. So we can get candidate set by abandoning most unrelated motion clips to reduce the number of costly similarity measure significantly. Experimental results show that our methods are effective for motion data retrieval in large database. © Springer-Verlag Berlin Heidelberg 2007.