基于关键姿态分析的运动图自动构建

高结点聚合运动图(snap together motion graph, STM graph)是刻画虚拟角色运动序列关系的一种结构化运动图.其特点是图中每个结点都包含多条与之相连的边，能够实现对虚拟角色的灵活控制.然而现有的高结点聚合运动图构建方法存在手工标注任务繁重、关键姿态提取结果不准确等问题.针对上述问题，提出了一种基于关键姿态分析的运动图自动构建新方法：通过维度约简和非参数密度估计分析样本数据的概率密度，获得一组关键姿态；然后通过分割获得运动片段，最后构建高结点聚合运动图.该方法不仅提高了关键姿态的提取精度，减少了构图过程的主观因素，同时提高了对虚拟角色控制的灵活性.实验结果表明了该方法的有效性.

Key-Postures Based Automated Construction of Motion Graph

STM graph (snap together motion graph), with high degree of polymerization nodes, is a structured graph to describe the relationship of the motion segments in character animation. Nodes in a motion graph serve as postures and edges between these nodes correspond to motion clips. Each node in an STM graph is connected with multiple edges. Many different approaches have been proposed to construct motion graphs from the existing motion capture data, which gives the user a flexible way to synthesize natural looking motion and control the character. So it becomes a hot topic to construct motion graphs automatically. However, the current methods of constructing STM graph depend largely on the experience and manual manipulations. Focusing on the problems mentioned above, a novel method to create motion graph automatically is proposed in this paper. Dimension reduction and nonparametric density estimation analysis are adopted to extract the key postures from motion capture data. The segments are obtained to construct the motion graph with high degree of polymerization nodes. The method not only improves the accuracy of extraction of key postures and reduces the subjective factor, but also improves the flexibility of controlling the virtual characters. Experiments have been done on taekwondo motion clip with 934 frames and badminton motion clip with 1798 frames. The results show the effectiveness of the method.