X-Conference : reinventing a teleconferencing system

Abstract

In looking forward to more natural we can anticipate that the teleconferencing system of the future will enable participants at distant locations to share the same virtual space. The visual object of each participant can be transmitted to the other sites and be rendered from an individual perspective. This thesis presents an effort, X-Conference, to reinvent a teleconferencing system toward the concept of "3-D Virtual Teleconferencing." Several aspects are explored. A multiple-camera calibration approach is implemented and is employed to effectively blend the real view and the virtual view. An individualized 3-D head object is built semi-automatically by mapping the real texture to the globally modified generic model. Head motion parameters are extracted from tracking artificial and/or facial features. Without using the articulation model, facial animation is partially achieved by using texture displacement. UDP/IP multicast and TCP/IP unicast are both utilized to implement the networking scheme.