Extraction of Feature Quantities Suitable for Distribution Visualization of Motion Capture Data

Abstract

To handle a motion-capture (Mocap) data set such as a Mocap database, it is beneficial to grasp the overview of the motion-characteristic distribution of the set in advance. To easily grasp the overview, concisely visualizing the distribution using a scatter plot is effective. In this paper, we propose a new method to extract the feature quantities suitable for the above visualization from each of the Mocap data. The one-dimensional motion-speed time series is analyzed in the frequency domain. Consequently, two feature quantities representing the motion intensity and motion complexity are derived. It is shown in the scatter-plot-construction experiments that explicitly weighting each frequency value in the frequency domain is effective for extracting the characteristics specific to each motion category.