Qifa Ke and Takeo Kanade December 2002
In Proceedings of the IEEE Workshop on Motion and Video Computing (MOTION)