Highly efficient facial blendshape animation with analytical dynamic deformations

Adding physics to facial blendshape animation is an active research topic. Existing physics-based approaches of facial blendshape animation are numerical, so they require special knowledge and skills, additional preprocess, large computer capacity, and expensive calculations leading to low animation frame rates, and are not easy to learn, implement and use. To tackle these problems, we propose an analytical approach and develop a blending force-based framework for physics-based facial animation. The proposed approach introduces the equation of motion to consider inertial effects, damping effects and the resistance against deformations, combines them with source and target facial shapes to formulate the mathematical model of dynamic deformations, and develops a simple and efficient closed-form solution. The blending force-based framework incorporates the new proposed slider force-based, exponentiation force-based and random force-based methods built on the obtained closed form solution to achieve highly efficient facial animation. Compared with facial blendshape animation using geometric linear interpolation, the proposed approach is physics-based. It not only creates all the blended shapes generated by linear interpolation, but also a much larger superset of blended shapes. Unlike linear interpolation which can only generate blended shapes with a same deformation rate, the proposed approach can generate blended shapes with different deformation rates, resulting in special effects of acceleration and deceleration. Compared to existing physics-based approaches of facial blendshape animation which are numerical, the proposed approach is the first time to develop an analytical approach of physics-based facial blendshapes. It does not require any special knowledge and skills and is easy to learn, implement and use. More importantly, it can avoid the additional preprocess of numerical methods and create various physics-based facial blendshape animations highly efficiently. Moreover, it can be used to estimate physical parameters from real shapes and developed into an interactive and real-time physics-based shape manipulation tool.

Duke Scholars

Author Feng Tian DKU Faculty