Çok-bakışlı video dizilerinden yüzey izleme

We present a fast and efficient surface tracking method for modeling dynamic objects from multiview video streams. Starting from an initial mesh representation, the surface of a dynamic object is tracked over time, both in geometry and connectivity, based on multiview silhouette and 3D scene flow information. The mesh representation of each frame is obtained by deforming the mesh representation of the previous frame towards the optimal surface defined by the time-varying multiview silhouette information, using mesh restructuring operations and vertex displacements assisted by 3D scene flow vectors. The whole time-varying surface is then represented as a mesh sequence that can efficiently be encoded in terms of restructuring operations and small-scale vertex displacements along with the initial model. Our reconstruction method hence yields a compact time-varying mesh representation of the dynamic object, which is smooth both in time and space. The proposed method is not only fast and produces storage efficient mesh representations, but it also has the ability to deal with dynamic objects that may undergo nonrigid transformation. The time-varying mesh structure of such nonrigid surfaces, which is not necessarily of fixed connectivity, can also successfully be tracked thanks to the restructuring operations employed in our deformation scheme. We demonstrate the performance of the proposed method both on real and synthetic sequences.