基于运动序列分割的运动捕获数据关键帧提取  

Extraction of Keyframe from Motion Capture Data Based on Motion Sequence Segmentation

作　　者：朱登明[1] 王兆其[1] 

机构地区：[1]中国科学院计算技术研究所虚拟现实技术实验室

出　　处：《计算机辅助设计与图形学学报》2008年第6期787-792,共6页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(60573162,60533070,60603082);国家“八六三”高技术研究发展计划(2006AA01Z336,2007AA01Z320);中国科学院“科技助残行动计划”项目(KGCX2-YW-610)

摘　　要：采用线性时不变系统把高维运动数据映射到低维状态空间;在低维状态空间中,定义了姿态之间的相似性度量;并采用误差平方和准则对时序的低维数据点集进行运动分割,分割点上的运动姿态被定义为关键帧.实验结果表明:该算法能够较好地提取出运动序列中的关键帧,并且这些关键帧能够很好地概括原始运动序列的内容.Linear time-invariant system is used to derive an explicit mapping between the high- dimensional motion capture data and the low-dimensional state variables. A similarity metric is defined to measure the difference between different poses in the low-dimensional state space, and then the method of mean squared error is employed to divide the motion sequence into a sequence of concatenated segments. The poses at these segmentation points are then defined as keyframes. Experimental results show that the extracted keyframes by our method can give a good visual summarization of the original motion sequence.

关 键 词：运动捕获数据 关键帧 线性时不变系统 运动分割 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]