A New Motion-Texture Graph for Generating Real-time Realistic Human Motions

Motion-Texture synthesis is a new field of research which aims at generating new human motions from a given motion capture (MoCap) data. The success of such techniques depends on the right selection of interpolation and distance functions among the motion data. Although the quaternion representation is the most mathematically accurate one, it is usually avoided due to the heavy computations needed. This paper proposes a fast technique for Motion-Texture synthesis using qauternions and based on Markov Random Fields Model (MRF) model. The current work proposes a new Motion-Texture Graph called Quaternions MRF Motion- Texture Graph (QMMTG). QMMTG technique uses quaternions to devise a proposed fast matching criterion and a proposed simple method for smoother transitions among QMMTG nodes. QMMTG technique manages to generate a wider range of new motions and even better results in some cases. Besides, the smoothing algorithm has perceptually enhanced the resulting motions in comparison to the previous works. The authors envision QMMTG as a start for benefiting from the quaternions accuracy in real-time Motion- Texture synthesis applications.