Manifold Latent Probabilistic Optimization for Human Motion Fitting Based on Orthogonal Subspace Searching
Wanyi Li · Journal of Information and Computational Science · 2014
This paper is aimed to propose a novel method called Manifold Latent Probabilistic Optimization based on Orthogonal Subspace Searching (MLPO-OSS). It can make Gaussian Process Dynamical Models (GPDM) learn incomplete human motion cycle to obtain regular latent variable data and estimate missing frames of the motion cycle well. Further more, other human motion cycle can be mapped from the incomplete human motion cycle via using the latent variable data to fit the sparse samples effectively, meanwhile, high dimensional human motion data fitting can be improved in this way. The validity of this method can be verified through the experiments.