Recovering 3D+t scenes from 3D and 2D+t data by motion compensation and using a prior surface model.

Wafa Rekik (1) , Dominique Béréziat (1) , Séverine Dubuisson (1)

1 PEQUAN - Performance et Qualité des Algorithmes Numériques

In this paper, we address 3D+t shape recovery from 3D spatial data and 2D+t temporal sequences. This reconstruction is particularly challenging due to the great deal of in-depth information loss observed on the 2D+t temporal sequence. To handle this critical lack of information, we use a geometrical local constraint defined by a spherical topology arising in number of computer vision applications. Our approach consists in gradual 2D-to-3D restoration by motion compensation using a brightness constancy constraint and a spatial regularity criterion. Data prior geometrical model implies that structures of interest evolve on a spherical support. Motion has then only two degrees of liberty. We build then an adapted 2D model in the surface domain, where move- ment assumptions are more relevant, using a spherical parametrization. We restore 2D+t surfaces parametrized by spherical angular coordinates that we display using an adapted visualization tool.

Mots clés: 3D+t reconstruction; optical flow; spherical parametrisation

Domaines: Traitement des images [eess.IV]

Wafa Rekik, Dominique Béréziat, Séverine Dubuisson. Recovering 3D+t scenes from 3D and 2D+t data by motion compensation and using a prior surface model.. 2010. ⟨hal-00512298⟩