Abstract
The automatic detection of the perceived origin of full-body human movement (OoM), i.e., of the part of the body that an external observer perceives as the joint where the movement originates, is a relevant topic for human movement analysis, as it can allow one to interpret affective content and social signals and can have applications in cognitive/motor rehabilitation, among others. Within this framework, the objective of this work is to present a computational method aimed at the automatic detection of the perceived OoM, starting from movement features acquired via motion capture techniques. After defining the concept of perceived OoM, the following contributions are presented: a set of techniques for the automated analysis of full-body expressive non-verbal communication, based on several low-level local movement features of the joints (speed, tangential acceleration, angular momentum); a computational method for the automatic detection of the perceived OoM at different spatial scales; and a repository of full-body movements annotated in terms of the perceived OoM, adopted for validation and evaluation of the method. The results of the analysis demonstrate its effectiveness. Finally, possible extensions of the method are outlined.