出版社：The Institute of Image Information and Television Engineers
摘要：Time-varying mesh (TVM), a technique that is used to describe the full shape and motion of a real-world moving object, precisely captures and reproduces human behavior and natural movements (such as the expression of a face or small changes in clothing). Unfortunately, TVM requires a large storage space and has a high computational cost. To solve this problem, we developed a framework of motion editing and synthesis aimed at rendering on mobile devices in particular, in which the computational resources are severely limited. In our approach, a skeleton model is extracted from each frame and motion tracking and analysis are conducted using the skeleton models. Then, a key frame is deformed based on the extracted motion information. This results in a 3D scene representation with a much smaller data size. The motion of the TVM model can also be controlled by using motion capture data. Experimental results showed that the data size can be reduced from 650 MB to 14 MB. In addition, subjective evaluation demonstrated that the proposed approach is appropriate for small-size display while maintaining a reasonable rendering quality.
