The ANR Human4D project aims to take on new challenges in the context of modeling the human body and to propose a new efficient 4D modeling of the human body in movement. Our ambition is to go beyond existing representations of shape space that primarily focus on static shape poses, and to consider dynamics.

The objective of this work is first of all to analyze movement sequences acquired on the Kinovis platform in Grenoble, sequences provided by the INRIA partner. It will first be necessary to look at existing methods from the literature which make it possible to extract relevant information (information which differentiates subject's identity from pose). We will be particularly interested in spectral analysis techniques applied to 3D shapes and will adapt them to take into account the temporal aspect of the data.

The second objective will be to propose new compact representations of the human body in movement by considering the properties of the shapes previously analyzed. These new representations should make it possible to extend recent learning techniques used on regular 2D grids to dynamic 3D entities. Their exploitation based on an analysis of moving shapes will allow a synthesis of new movements/new forms.