Mesh skinning has been a widely applied method in games for skeleton driven character animation. A gaming character can be easily animated and deformed by transforming every vertex using a weighted sum of proxy bone transformations in the skeleton. Arbitrary deformations such as facial and cloth animation can also be effectively represented with a relatively large number of proxy bones. However, thelarge number of necessary proxy bones make it difficult to intuitively control the animation of deformable surfaces. This dissertation investigates novel methods for data-driven deformation and level-of-detail rendering of skinned mesh.The first part of this dissertation includes methods to effectively generate new animation for a skinned mesh, which may be a non-articulated model or highly deformable surface such as face or clothes. We develop a regression framework to learn deformation styles for a skinned mesh from example configurations based on kernel Canonical Correlation Analysis (CCA). Without expensive non-linear optimization at run-time stage, the regression method is very efficient and can achieve real-time performance in generating novel mesh deformations. The regression method is also investigated to generate deformation details for dynamic clothes. A hybrid approach for cloth animation is developed to find a mapping between coarse deformations to high-resolution spatial details in a cloth model. The quality of regression model is improved by making separate regressions at different detail scale and by identifying suitable rotation-invariant quantities for regression. The run-time components are implemented efficiently on GPU to achieve an overall real-time performance on high-resolution cloth models. In the second part, we focus on real-time rendering methods for skinned mesh. We introduce feature preserving triangular geometry images for level-of-detail rendering of skinned mesh. Triangular charts pack efficiently, simplify the elimination of T-junctions, arise naturally from an edge-collapse simplification base mesh, and layout more flexibly to allow their edges to follow curvilinear mesh features. By incorporating skinning weights and skinned bounding boxes into the representation, a view-dependent LOD scheme can be applied for rendering skinned meshes stored and rendered entirely on the GPU to maximize throughput. We also develop a data management scheme for precomputed radiance transfer to render skinned models with global shading effects.
展开▼
