Local non-rigidity inference

Volumetric correspondences with random forests

My research is concerned with deformable motion of 3D shapes acquired from multi-camera environments such as the kinovis platform at Inria Grenoble. I design algorithms that simultaneously track shapes and learn deformation properties. This combination allows for better shape tracking and it provides knowledge about the observed shapes dynamics. My latest work aims at predicting volumetric correspondences using supervised machine learning.

I work with both surface-based and volume-based representations of shapes, thanks to centroidal Voronoi tessellations which are quasi-regular exact shape tessellations.

The main challenge with 3D reconstructed sequences is to handle data noise and the temporal uncoherence of meshes reconstructed at different timeframes.


keywords: shape tracking, deformation model, 3D registration, temporal coherence, centroidal Voronoi tessellation