Enclosing Mesh Generation And Mesh Sequence Reusing

With rapid development of advanced 3D acquisition techniques, high-resolution 3D models and animated mesh sequences are booming. It is an urgent task to reuse these digital geometry resources effectively and efficiently in many areas, including comput-er animation,3D computer games and virtual reality, etc. In cage-based deformations, a coarse enclosing mesh called cage, as a proxy of the embedding geometry, has the advantages of decoupling related algorithms from representation and complexity of the underlying model. Thus, cage-based deformation algorithms are universal and efficien-t. However, current cage generation methods are related with geometric representation and complexity of the model in general, which narrows down the scope of cage-based methods. For reusing a mesh sequence, it is prevalent to construct a cage sequence en-coding the mesh-based animation. However, the cage sequence is determined solely by the geometry and topology information of mesh sequence, which makes it difficult to in-corporate user’s interactive intentions for animation processing. Considering that a mesh sequence contains two aspects, i.e., motion and geometry, existing motion blending or static shape morphing techniques cannot exploit the potentials of the mesh sequence. Thus, the problem of reusing two mesh sequences are still to be explored. Therefore, some novel methods are proposed to address the above mentioned problems from the as-pects of cage generation and cage-based mesh sequence reusing. The main contributions of the thesis include:To address the lack of generality of current cage generation methods, a cage gener-ation algorithm based on visual hull is proposed. The 3D reconstruction technique based on image information in computer vision is introduced to avoid relying on the geomet-ric information of the underlying model by converting the 3D cage modeling into 2D image modeling. Thus, the cage generation can be independent of geometric represen-tation and complexity of the underlying model. As a result, the proposed method can be applied to arbitrary shape representation in case the model can be rendered as a series of images. Additionally, the proposed method offers the means of adjustable cage tightness and local cage construction for the model.To address the lack of high-level control means in cage-based representations, an adaptive skeleton-driven cage is proposed for mesh sequence representation and reusing. The main idea of the proposed method is to combine the strengths of the skeleton- and cage-based representations by introducing the skeleton to help adaptively generate an initial cage to preserve time-varying geometric details and using the skeleton as an intu-itive interface for animation processing. To robustly extract an kinematic skeleton for the reference pose, the topology of the skeleton is inferred by sketching on the reference pose mesh, and then the joint positions are determined by computing prominent cross-sections at user-specified points. On the other hand, the initial cage is adaptively constructed un-der the guidance of these cross-sections and the cage vertices are bound to the joints of the corresponding cross-sections. The propagation of the initial skeleton-driven cage to the mesh sequence can be dealt with two independent problems. The skeleton se-quence is extracted by propagating the initial skeleton to the other mesh based on the temporal coherency of the mesh sequence. The cage sequence is computed efficiently by solving a pre-factorization least-squared fitting problem. Additionally, the generated cage sequence can be adaptively refined to improve the mesh reconstruction quality con-trolled by the reconstruction error. The implementation results and detailed comparisons demonstrate that the proposed approach is an intuitive, compact, and efficient cage-based representation. We demonstrate its potentials through a variety of applications, such as animation compression, deformation transfer, pose editing, etc.For the reusability between mesh sequences, a novel morphing concept for mesh sequences is proposed, i.e., generating an intermediate sequence with time-varying mor-phing shapes and blended motion style from two different sequences. To fully make use of geometry and motion information in mesh sequences, a universal morphing frame- work is devised that is applicable to all possible combinations of motion blending and dynamic shape interpolation. The proposed framework consists of four key techniques: mesh sequence representation, cross-parameterization, motion blending and dynamic shape interpolation. Considering that the proposed adaptive skeleton-driven cage has the advantages of skeleton and cage, two input mesh sequences are encoded using this control structure to blend motions by skeleton and capture the geometry by cage. To enable a universal framework for various morphing combinations, a skeleton-driven cage-based deformation transfer scheme is elaborately designed, which can account for motion blending and geometry interpolation. To minimize user’s effort, a hybrid cross-parameterization scheme is introduced by combining the domain-based parameterization and template-based fitting approach to establish one-to-one correspondence for inter-polation between two mesh sequences. The experimental results demonstrate that the framework, not only accomplishes mesh sequence morphing, but also is suitable for a wide range of applications, such as deformation transfer, motion blending or transition, and dynamic shape interpolation.