Optimizing orthonormal basis bilinear spatiotemporal representation for motion data

Abstract

The paper describes an attempt to estimate the optimal division of a number of base vectors between space (shape) and time (trajectory) for bilinear spatiotemporal representation ofmotion capture data. The spatiotemporal model is amatrix consisting of Ks · Kt amount of coefficients. In the paper we discuss using of orthonormal spatial and temporal basis: PCA-PCA, DCT-DCT, PCA-DCT and DCT-PCA to represent real MoCap data.