A hybrid P/KPCA-based approach for motion capture data automatic segmentation

Abstract

A Motion Data Automatic Segmentation using a Probabilistic/Kernel principal component analysis (P/KPCA) method is proposed. This approach utilizes Kernel principal component analysis (KPCA) to construct a kernel function while using Probabilistic principal component analysis (PPCA) to reduce motion noise. Formulate the feature function to obtain the derivative of projection error, and detect the segmentation point of data through analyzing the change of geometric features to realize the automatic segmentation. It is indicated in the experiment that the motion capture technique has certain feasibility. The paper presents the automatic segmentation approach of the motion capture data, in which the motion data is automatic segmented through KPCA combined with PPCA to reduce the dimension and project the 56 dimensional data in 2 dimensional space; formulate the feature function to obtain the derivative of projection error, and detect the segmentation point of data through analyzing the change of geometric features to realize the automatic segmentation. It is indicated in the experiment that the motion capture technique has certain feasibility.