Keyframe Extraction from Motion Capture Data for Visualization

In this paper, we propose a novel method to extract keyframes from motion capture data for people to better visualize and understand the content of the motion. It first applies a Butterworth filter to remove the noise in the motion capture data, then carries out principal component analysis (PCA) to reduce the dimension. By detecting the zero-crossing points of the velocity in the principal components, the initial set of keyframes can be derived. The initial set of keyframes are further pruned to group similar postures together to remove redundancies. An experiment based on the CMU motion capture database is carried out, and the experiment result suggests that the keyframes extracted with our method could better support visualization and understanding of motion capture data.