动态几何序列生成与压缩研究

发布时间：2018-03-22 06:27

本文选题：数字几何处理　切入点：形状插值　出处：《华南理工大学》2016年博士论文　论文类型：学位论文

【摘要】：基于网格的动态几何序列生成与压缩是计算机动画的理论基础.几何序列的生成方式主要有扫描仪获取、基于关键帧的形状插值、基于物理驱动的变形、对已有几何序列进行编辑与操纵等.形状插值因其交互少、控制简易以及效果直观,成为动画设计常用的几何序列生成技术.通过对已有几何序列进行编辑与操纵,动画设计师能更加灵活地获得更丰富的、更具个性化的动态几何序列.另一方面,生成的原始几何序列往往占用大量存储空间,需要压缩技术对其进行高效存储.基于视觉误差的压缩方法在近年备受关注.本文对基于三维网格的形状插值、序列编辑、序列姿态移植以及序列压缩等内容进行了研究.本文提出一种快速拟等距三维网格插值技术,将形状插值问题表示为寻找三角形边向量的拟等距路径的非线性问题,并建立相关数学模型.求解非线性问题的初始化阶段,提出一种基于边正交标架及连结映射的传播-优化算法,能有效、快速地解决大尺度变形插值问题;在迭代优化阶段,采用块坐标下降法,将序列的全体未知量优化解耦为逐条边的优化,并进一步简化后者的求解,显著降低计算复杂度.对于序列编辑与姿态移植,通过操纵每一帧邻接顶点的相对速度以达到修改序列模型姿态的目的.最后,提出一个适用于形状插值、序列编辑以及序列姿态移植的统一计算框架.实验表明,相比目前最好的形状插值技术,拟等距形状插值算法能够获得更加符合等距约束的序列,而且更加高效.本文提出一种新的同构网格序列的压缩算法.注意到对齐序列中的姿态会增加信息冗余度,本算法分析由网格序列所构成的顶点轨迹矩阵,通过刚性变换对齐序列模型,寻找轨迹子空间,使轨迹的线性相关性尽可能大.首先,算法对整个序列进行运动分析并分割网格模型的刚性块,然后对顶点轨迹矩阵进行低秩分解,用以引导刚性块对齐,并采用主成份分析表示刚性块对齐后的轨迹.刚性变换、主成分分析的基与其混合系数使用现有动态网格压缩技术保存.实验结果表明,在相同的视觉误差条件下,本算法的性能与目前最好的压缩方法相当;如果序列的运动接近分块刚性,本方法具有较大的压缩优势.
[Abstract]:Generation and compression of dynamic geometric sequences based on mesh is the theoretical basis of computer animation. Geometric sequences are generated by scanners, shape interpolation based on key frames, and deformation based on physical drive. Shape interpolation has become the common geometric sequence generation technology in animation design because of its less interaction, simple control and visual effect. By editing and manipulating the existing geometric sequence, the shape interpolation has been used in animation design. Animation designers are more flexible in getting richer, more personalized dynamic geometric sequences. On the other hand, the original geometric sequences often take up a lot of storage space. The compression method based on visual error has attracted much attention in recent years. In this paper, shape interpolation based on 3D mesh, sequence editing, In this paper, a fast quasi-equidistant three-dimensional mesh interpolation technique is proposed, in which the shape interpolation problem is expressed as the nonlinear problem of finding the quasi-equidistant path of the triangular edge vector. At the initialization stage of solving nonlinear problem, a propagation-optimization algorithm based on edge orthogonal frame and link mapping is proposed, which can effectively and quickly solve the large-scale deformation interpolation problem. By using the block coordinate descent method, the optimization of all unknown variables of the sequence is decoupled to the edge-by-side optimization, and the solution of the latter is further simplified, and the computational complexity is significantly reduced. By manipulating the relative velocity of the adjacent vertices of each frame to modify the posture of the sequence model, a unified computing framework suitable for shape interpolation, sequence editing and sequence attitude transplantation is proposed. Compared with the best shape interpolation technique at present, the pseudo-equidistant shape interpolation algorithm can obtain the sequence which is more consistent with the equidistant constraint. In this paper, a new compression algorithm for isomorphic mesh sequences is proposed. It is noted that the pose of alignment sequences increases the redundancy of information. The algorithm analyzes the vertex locus matrix composed of mesh sequences. The linear correlation of the trajectory is as large as possible by aligning the sequence model with the rigid transformation. Firstly, the algorithm analyzes the motion of the whole sequence and segments the rigid block of the mesh model. Then the vertex locus matrix is decomposed in low rank to guide the alignment of rigid blocks, and principal component analysis (PCA) is used to express the trajectory after alignment. The basis of principal component analysis and its mixing coefficients are preserved by existing dynamic mesh compression techniques. The experimental results show that the performance of this algorithm is comparable to that of the best compression method under the same visual error conditions. If the motion of the sequence is close to block rigidity, this method has a large compression advantage.
【学位授予单位】：华南理工大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP391.7

【相似文献】