基于分块主成分分析的人体运动合成

摘要

在高维运动数据处理中，传统降维方法过度关注于保护拓扑结构和重构信息，导致低维子空间参数具有不可理解性。针对该问题，提出一种基于分块主成分分析的运动合成方法。对根据人体骨骼结构特点分组的运动数据进行分块主成分分析，获得具有特定语义的低维子空间参数。以跳跃运动为例进行实验，结果表明，该方法通过直观地改变具有语义的运动参数，可实时合成满足要求的运动结果。%Human motion analysis based on traditional dimensional reduction methods often overstresses topology preservation and reconstruction precision which results in incomprehensible low-dimensional subspaces. To solve this problem, this paper proposes a Block Principal Component Analysis(BPCA) based method to analyze motion capture data and further synthesize human motions interactively. By applying BPCA to different body parts, low-dimensional motion parameters which are semantically meaningful can be obtained. Taking jumping motion as the example, experimental results show that by intuitively adjusting these motion parameters, desired new motions can be generated in real-time.