Human-Robot Interaction: Invariant 3-D Features for Laban Movement Analysis Shape Component

L. Santos and J. Dias (Portugal)


Human-Machine Interaction, Laban Movement Analysis, Bayesian Models, Gesture Recognition, Signal Processing.


In the field of human-machine interaction, there are still lacking efficient tools within visual perception of human non-verbal comunication. In this context, some investigation has been conducted within the paradigm of Laban Movement Analysis (LMA) [1–5]. This work, will explore how visual signals can be processed in order to retrieve useful features which will allow the characterization of movements in a semantic/intuitive way (e.g. reaching). The LMA shape component will be the main focus of this work, and the implementation will follow the guidelines of previous work. The head and both hands will be tracked within the image, and stereo vision model will be used to retrieve 3-D information of the performer’s pose. From this 3-D data, invariant features will be generated, and used as evidences in a Bayesian framework, which is the selected tool for Laban Movement Analysis implementation. The current work intents to further extend the LMA Bayesian models, towards a full robust descriptor of non-verbal cues for machine interpretation of human behaviour. Results show that the more complete the global Laban Movement Anal ysis model becomes, better results are achieved, leading to the thought that Laban can provide a good computational movement classifier.

Important Links:

Go Back