| dc.description.abstract | A technique is presented that permits fast view-reconstruction of individual objects. This method improves a previous approach to solve the problem of approximated view reconstruction by combining clustering of polygons with visibility bitfields to determine visibility for novel viewpoints. The technique consists of three steps: patch creation, spherical sampling, and rendering. In the first stage, the input 3D model is tiled in polygonal patches. In the sampling stage images of the model are taken from several points on the surface of a viewing sphere. Patch-ID bitfields, which are structures that contain visibility information, are computed for each picture. In the rendering stage, a subset of the viewpoints computed for sampling is selected depending on the viewers position on the viewing sphere and the bitfields of the selected viewpoints are used to rebuild the visible parts of the model from the new viewpoint. The overall system presented here makes a very efficient use of memory resources, and involves practically no overhead during rendering while significantly improving frame rate during interaction with large models. Although the technique is not conservative, our results show that the reconstructed views are practically identical to the original views of the model. | en_US |
