Abstract
3D video is the ultimate image medium recording a dynamic visual event in the real world as is. The recorded object can be observed from any viewpoint, because 3D video records an object's full 3D shape and its surface properties(i.e.color and texture). Recently, methods of creating arbitrary object views from multi-viewpoint images have been proposed. Although some methods do not reconstruct an object's 3D shape, we believe the explicit representation of the 3D shape is required for object action recognition and versatile editing of multiple 3D videos. This paper presents methods of 1) transforming voxel representation of an object into polygonal representation, 2) generating texture on the surface patches from multi-viewpoint images, 3) visualizing 3D video with versatile camera works and object-background configuration.
