Home Publications
Home Publications
The informative content of optical flow features of biological motion
Year: 2014  
Authors: Alessandra Sciutti, Nicoletta Noceti, Francesco Rea, Francesca Odone, Alessandro Verri, Giulio Sandini  
Volume: 43
Book title: ECVP Abstract Supplement
Series: Perception Pages: 102
A natural predisposition towards biological motion, i.e. motion produced by a living being, is shown in human neonates since the earliest stage of the development. An innate ability allows them to perceive the presence of other humans moving in their social space, even with the limited visual information available at birth. In this work we show that a perspective invariant visual feature, computed from optical flow, could be used to explain this ability and we test this hypothesis through action observation and using a humanoid robot engaged in a human-robot cooperative task. To assess the validity of our idea we record videos of non-biological motions (e.g., bouncing ball, pendulum) and biological motion (e.g., human actions, as lifting and transporting objects) and we show that it is possible to classify movements in a scene as biological or not just on the basis of dynamic features computed from optic flow. Moreover, we demonstrate that it is possible to extract subtle action properties (e.g., the weight of the object lifted) from these features. The derived model is then implemented on a humanoid robot, to test the validity of our assumptions in real human-robot interactive scenarios.