Floating 3D video conference

Abstract

This paper proposes an improved algorithm based on Active Appearance Models (AAM) and applies on a real-time 3D video conference system with a novel 3D display device which can pop out an avatar out of the display in the air. The proposed algorithm utilizes an improved Adaboost algorithm [1] for face detection based on skin color information. Facial feature points are then tracked based on AAM [2] and we improved the algorithm to determine the rate of closing eyelid and the rotation angle of eyeballs. The novel 3D display device projects digital images on an actual human like object as an avatar and pops out a 3D image in the air via an optical module. With the proposed system, users can interactive intuitively with a popped 3D avatar. This system provides more realistic and representative visual effect for interaction in a video conference. © 2011 Springer-Verlag.