Movement sequences are essential to dance and expressive movement practice; yet, they remain underexplored in movement and computing research, where the focus on short gestures prevails. We propose a method for movement sequence analysis based on motion trajectory synthesis with Hidden Markov Models.
Our analysis method draws upon the synthesis of motion trajectories and their associated confidence interval from a probabilistic model trained with several executions of a movement sequence. We use Hierarchical Hidden Markov Regression for motion trajectory synthesis. We showed that investigating the confidence intervals reveals variations in consistency along the performance, and can serve as basis for analyzing consistency both within and between performers. While their primary use has been gesture recognition, probabilistic sequence models such as HMMs are promising for movement sequence analysis. The main advantage of the method is the possibility to learn from several examples and to account for uncertainty in movement execution and capture.
We illustrate the method with a case study in Tai Chi performance. Our results reveal that the performers’ variability evolves over time, and is minimized for a set of specific gestures, especially as the performer’s expertise increases. We extended the method to the cross-modal analysis of vocalized movements, highlighting that vocalizations are consistently performed on the gestures executed with high accuracy, often highlighting metaphorical aspects of the performance.
This project has been conducted with Agnès Roby-Brami and Natasha Riboud.