Multiscale Processing Performance for Motion Capture

Motion capture systems help record human motion as a sequence of joint angle vectors and analyse it in multiple degrees of freedom with high accuracy. Motion, as many other signals, might contain information which is stored on many different scales. Hence the use of a multiscale model might help correctly distinguish or analyse motion properties. In this paper we analyse the capabilities of a multiscale motion model to help distinguish meaningful motion features, whilst the unnecessary components (like noise) get removed. We performed experiments based on real motion capture data to analyse the discriminative properties of the multiscale approach. The main goal of experiments was to check the clustering performance of a multiscale model. The detailed results are presented and discussed, showing the capabilities and advantages of multiscale model application.