Animation and Performance Capture Using Digitized Models

Více o knize

The generation of realistic virtual doubles of real-world actors has been a long-standing challenge in computer graphics, with several issues still unresolved. Traditional skeleton-based character animation remains time-consuming, passive performance capture of actors in everyday clothing poses difficulties, and there are limited techniques for processing mesh animations compared to the abundance of skeleton-based methods. This thesis addresses these challenges with algorithmic solutions. It introduces two efficient mesh-based alternatives that simplify character animation while maintaining realistic body deformations, compatible with traditional pipelines. Additionally, three passive performance capture methods are developed, utilizing a deformable model for scene representation, enabling the reconstruction of coherent time-varying geometry, motion, and texture for subjects in loose apparel. These high-quality reconstructions facilitate the rendering of realistic 3D videos. Furthermore, two novel algorithms for mesh animation processing are presented: one for automatic conversion to skeleton-based animation and another for creating an animation collage, a new artistic rendering style. Collectively, these methods provide solutions to specific problems and serve as foundational elements for a comprehensive system that captures, manipulates, and realistically renders human performances, surpassing many existing techniques. This