Gaussian Process Regression for Animation Rig Towards the Face Model

In professional 3D animation artists model movements and scenes using rig functions - constrained set of sliders or controllers that propagate deformations and drive mechanism of object or character in systems of 3D tools. These controllers are manually built for each character and cannot be reused if the underlying structure is not exactly the same. There are often hundreds of adjustable parameters, and artists have to learn the structure for each new character. This is usually bottleneck in production, that might be avoided by automating this process. D. Holden et. al. proposed possible solutions using Gaussian Processes Regression, which showed useful in the case of skeletal (quadriped) characters. We want to further apply this on face model, that has a completely different structure than the skeletal model. In this work we explain the model for 3D face animation, the theory of Gaussian processes regression and a method to apply it for solving the problem of interest. At the end results and examples are presented with a simple animation model we have at our disposal.