Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE credits
This thesis describes two tools for facial animation that could improve the rigging and animation workflow at Moving Picture Company, MPC. The motivation was that it is very difficult to modify or re-use work made within these departments, and new tools could save time from tedious and repetitive work.
Two specific problems were adressed: vector propagation in order to reconstruct partial blendshapes, and shape transfer in order to re-use facial expression between different meshes.
A survey over related work was conducted in order to find appropriate methods. Global methods such as Laplace Surface Editing, and Global Shape transfer were considered the best methods to solve each problem respectively. However, due to constraints in the development environment, the problems were solved using alternative methods.
The tools were implemented in KL using Splice Maya editor and Fabric Engine. They were tested in Maya, but are portable due to using Fabric Engine.
The vector propagation tool was implemented using a linear per-vertex approach that computes forces in relation to a deformed area. The results were good for certain situations, but too slow, difficult to use and unreliable to be truly useful.
The shape transfer tool was implemented as UV expression cloning. Motion vectors that describes a facial expression were transferred to a a second mesh through overlapping UV-space. The motion vectors were then corrected in magnitude and direction to account for differences between the meshes.
The shape transfer tool worked well for similar meshes, and was particularly effect for transfer between high and low-resolution meshes. It was fast-enough to transfer animation at interactive rates, but required overlapping UVs and did not achieve good results for characters with large differences.
Future work could be to either improve both tools, investigate alternative methods (particularly for vector propagation) or to develop a framework that could use both these tools and create a ”cut-copy-paste” framework for animation.
2016. , p. 45