3D sequence generation from multiple cameras

An approach for the generation of 3D surface model sequences from a set of synchronised, calibrated cameras is presented. The method is based on the computation of the visual-hull. An analysis classifies typical error sources for artefacts in the reconstructed 3D models into occlusion, approximation and sampling errors. For the synthesis of new images for TV- and film-applications the roughness of the reconstructed 3D models has shown to be very disturbing. In order to reduce this roughness two methods are used sequentially: a new super-sampling approach that reduces the sampling error, and a Gaussian surface smoothing. The application of the modelling results from a 12 camera studio system is shown for the generation of virtual views of moving actors in standard post-production packages.