Flot d'occupation 3D ` a partir de silhouettes latentes

In this paper we investigate shape and motion retrieval in the context of multi-camera systems and we propose a new low-level analysis based on latent silhouette cues. Many shape and motion analysis tools rely on the use of explicit surface models. Our analysis does not rely on explicit surface boundaries and uses an EM framework to simultaneously retrieve a set of volumetric voxel occupancy probabilities and a best estimate of the dense 3D motion field from the last consecutively observed multi-view frame set. As the framework uses only latent, probabilistic silhouette information, the method yields a promising 3D scene analysis tool robust to many sources of noise in difficult lighting and outdoor conditions. It can be used as input for higher level shape modeling and structural inference tasks. We demonstrate its practical use for shape and motion analysis experimentally.