Skeletal character animation from examples

Abstract

Accesses to 3D character models are becoming more and more handy and affordable through 3D scanning, reconstruction, artistic sculpting, etc. Animating these characters to tell a story opens up many applications in educations, communications, scientific research and entertainments. Current procedure of character animations requires considerable amount of labor and time dedicated by many skilled and talented artists. Exploring potentials in a character shape together with a set of example meshes in varied poses to enhance the animation methodology will definitely help in producing animations efficiently. However, most of those examples are pure geometry meshes and lack sufficient topology and pose information, which are critical in producing faithful animations. Meanwhile many available example meshes are too complex to use directly for animation production and associated computing demands heavy memory and CPU load. To tackle the above problems, in this dissertation, a set of methods is proposed and investigated to facilitate animation by exploring the potentials of example meshes. These methods, as the results demonstrate, provide an effective pipeline for generating faithful, visually pleasing and pose consistent animations. In this dissertation, first, a novel method on skeletonization from examples using harmonic one-forms is proposed. Extracting skeletons from 3D objects is challenging, especially for finding exact centered skeletons. The proposed method is motivated by the observation that many real-world deformations are isometric or near isometric from the global point of view. Compared to existing skeletonization methods, using harmonic oneforms bears no restriction on the connectivity of the input meshes, which is a common situation when a character is digitized in varied poses. Second, a method of pose parameterization of given example meshes is proposed. Pose parameters are suggested in this dissertation as Euler angles and scaling values for skeletal i ATTENTION: The Singapore Copyright Act applies to the use of this document. Nanyang Technological University Library I'm indebted to my internship mentor John Lewis on the idea of inverse PSD. His guidance in my coding skill and research methods are invaluable to my progress. I also thank Dr. He Ying for his support and guidance in this research, especially his idea and suggestions on skeletonization using harmonic 1-forms and solving the optimal joints in example poses. I also want to thank Prof. Sylvain Brandel from who I learned great programming style. Deep thanks also go to Dr.