Abstract
This paper approaches the problem of compressing temporally consistent 3D video mesh sequences with the aim of reducing the storage cost. We present an evaluation of compression techniques which apply Principal Component Analysis to the representation of the mesh in different domain spaces, and demonstrate the applicability of mesh deformation algorithms for compression purposes. A novel layered mesh representation is introduced for compression of 3D video sequences with an underlying articulated motion, such as a person with loose clothing. Comparative evaluation on captured mesh sequences of people demonstrates that this representation achieves a significant improvement in compression compared to previous techniques. Results show a compression ratio of 8-15 for an RMS error of less than 5mm.