Movement Analysis and Decomposition with the Continuous Wavelet Transform.

Human movements support communication, and can be used to imitate actions or physical phenomenons. Observing gestural imitations of short sounds, we found that such gestures can be categorized by their frequency content. To analyse such movements, we propose an analysis method based on wavelet analysis for clustering or recognizing movement characteristics. Our technique draws upon the continuous wavelet transform to derive a time-frequency representation of movement information. We propose several global descriptors based on statistical descriptors, frequency tracking, or non-negative matrix factorization, that can be used for recognition or clustering to highlight relevant movement qualities. Additionally, we propose a real-time implementation of the continuous wavelet transform based on a set of approximations, that enables its use in interactive applications. Our method is evaluated on a database of gestures co-executed with vocal imitations of recorded sounds.