Learning the Alphabet of Rhythmic Motion

Movement recorded from motion capture can be decomposed along two separate dimensions. It can be divided into shorter units of motion with respect to time. Secondly, a movement can also be decomposed into a base motion and a motion texture. We define the term, basic movement, to reflect properties from both of these decompositions of movement; it adheres to the atomic length property in time and also describes the baseline graph of a motion. We additionally define the terms, alphabet, as the complete set of all distinct basic movements in a motion database, and grammar rules as the set of all frequently occurring sequences of the basic movements. Determining the alphabet and the grammar of a motion capture database assists in addressing the organizational issues plaguing the process of motion capture, as well as provides a notational representation of motion. In this thesis, we introduce a method for determining the alphabet and grammar of rhythmic motion. Thesis Supervisor: Jovan Popović Title: Assistant Professor, MIT Laboratory for Computer Science