Using Subdivision Surfaces for 3–D Reconstruction from Noisy Data

In this work, we developed a method which effectively uses subdivision surfaces to fit deformable models to noisy 3–D data. The subdivision models provide a compact low dimensional representation of model parameter space. They also allow a free form deformation of the objects of arbitrary geometry. These two properties make them ideal for fitting deformable models to noisy data. We tested our approach on stereo data acquired from uncalibrated monocular video sequences, then on data acquired from low quality laser scans, and compared them. We have shown that it can be successfully used to reconstruct human faces as well as the other 3D objects of arbitrary geometry.