Evidential Reconstruction of a Sportive Gesture from a Multisensor Acquisition

We describe a reconstruction method of three dimensional marker trajectories from several cameras in the applicative framework of sportive gesture analysis. The principle is to apply a tracking algorithm on extracted markers from each sequence of images, then to fuse data to reconstruct three dimensional space movements of an athlete. The measurement-to-track association algorithm is based on the formalism of Dempster-Shafer's Theory of Evidence in the frame of the extended open-world, a powerful tool to represent uncertainty. Finally, the fusion enables to obtain a system which becomes efficient in its capacity to detect either the appearance or the disappearance of markers