Example-based dynamic deformation

• Main Author: Zhang, Wenjing
• Other Authors: Zheng Jianmin
• Format: Theses and Dissertations
• Language: English
• Published: 2016
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Computer graphics
• Online Access: http://hdl.handle.net/10356/65894
• Institution: Nanyang Technological University

This research investigates computational models and algorithms for efficiently incorporating example-based techniques into traditional physically-based deformable models to generate example-based dynamic deformation. Simulating deformable model plays a very important role in computer graphics and animation. While physically-based deformation provides highly realistic effects, it faces difficulties in high-level artistic deformation required in certain areas like computer animations. On the other hand, example-based techniques provide intuitive control for the deformation results by designing examples, which easily represent the high-level intention about the desired deformation. However, most of the existing example-based techniques focus on static deformation and interesting dynamics can be hardly captured. Thus, it is tempting to combine physically-based deformation techniques and example-based techniques for modelling dynamic deformation. Previous attempts addressing the problem either suffer from computational burden or sacrifice physical plausibility. The objectives of our research are thus to develop new computational models and algorithms that provide the user intuitive example-based approaches to control physically-based deformation, reduce the computation complexity and meanwhile preserve the physical plausibility as much as possible. Firstly, we present a new algorithm to simulate example-based elastic solid in real-time. The main idea is to make the simulation subspace integration friendly. For this purpose, we formulate an example-based potential using example-based Green strain tensors. With this new potential, both the reduced conventional internal force and additional internal force induced by examples are cubic polynomials in reduced coordinates. The subspace integration can then be developed with integration costs independent of geometric complexity. This greatly decreases the time complexity and allows real-time simulation. Secondly, we develop an algorithm for construction of inversion free and topology compatible tetrahedral meshes. This is useful for generating valid example poses for our example-based elastic solid simulation, which requires the tetrahedral meshes of example poses to be inversion free and topology compatible to that of the rest pose. We propose to warp the original tetrahedral mesh guided by the deformation of its boundary surface. The algorithm is devised to combine radial basis function (RBF)-based warping and adaptive mesh refinement. We iteratively transform the mesh using RBF-based warping with a safe stepsize to ensure that no element is inverted. To avoid too small stepsize, we refine the elements that are potentially inverted. The refinement is performed on the original and the warped meshes in the same way so as to maintain compatible topology between them. Thirdly, we propose a new method to efficiently animate solid or shell objects with both forward simulation and spacetime constraints in an example-based fashion. Our method directly extends the linear FEM. We devise a new example-based potential, which is quadratic in the deformation. Although less physically plausible than our previous example-based solid, this new potential allows the dynamic equations to be decoupled into single vibrations through modal analysis. Modal warping can be directly adopted to deal with large deformation. Due to the decoupling of the motion equations, either forward simulation or spacetime constraints can be solved very efficiently. Finally, we propose an efficient method to estimate the elastic parameters of a solid object based on a set of deformation samples of the object. We formulate the problem of estimating the material parameters into an optimization problem. A direct optimization is usually too time-consuming. We propose a method combining a new material model reduction and a new shape model reduction to address the limitations of previous methods. Particularly, in a sharp departure from previous shape model reduction techniques, we build the reduced model away from the current deformation rather than the rest pose. Our method can speed up the optimization and also enhance the accuracy. To efficiently enforcing box constraints for valid parameters (e.g. nonnegative Young's modulus), unlike previous material model reduction, we only reduce the search direction of the optimization to the subspace with material model reduction while keeping the projection onto constraints in full-rank space. Our method is much more efficient and less likely to get stuck during optimization.