Wavelet Statistics for Human Motion Classification

Human motion is as much characterized by its low frequency shape as by its high frequency temporal discontinuities – such as when a joint reaches its physical limit or when a foot touches the floor. Wavelets are particularly efficient at capturing both high and low frequency information. We introduce a method of classifying human motion using wavelet coefficients to build a representation of human motion signals. The representation is computed by finding the histograms of the wavelet coefficients previously scaled according to frequency. We use Support Vector Machines (SVMs) to classify those histograms and demonstrate the accuracy of the method on human motion gathered from both a motion capture systems and accelerometers.

Authors

• Kevin Quennesson Elias Ioup

Keywords

No keywords are indexed for this paper.