Free-viewpoint video of human actors

Abstract

In free-viewpoint video, the viewer can interactively choose his\n viewpoint in \\mbox{3-D} space to observe the action of a dynamic\n real-world scene from arbitrary perspectives.\n\n The human body and its motion plays a central role in most visual media\n and its structure can be exploited for robust motion estimation and efficient \n visualization. This paper describes a system that uses multi-view\n synchronized video footage of an actor's performance to\n estimate motion parameters and to interactively re-render the\n actor's appearance from any viewpoint.\n\n \n The actor's silhouettes are extracted from\n synchronized video frames via background\n segmentation and then used to determine a sequence of poses\n for a \\mbox{3D} human body model.\n By employing multi-view texturing during rendering, time-dependent changes \n in the body surface are reproduced in high detail.\n The motion capture subsystem runs offline, is non-intrusive, yields robust \n motion parameter estimates, and can cope with a\n broad range of motion.\n The rendering subsystem runs at real-time frame rates using ubiquous\n graphics hardware, yielding a highly naturalistic impression of the\n actor.\n The actor can be placed in virtual environments to create composite dynamic \n scenes. Free-viewpoint video allows the creation of camera fly-throughs or \n viewing the action interactively from arbitrary perspectives.