Current Issue Cover
基于高斯过程动态模型的人体节奏运动合成

吕培,张明敏,徐明亮,李灵,宋青见,潘志庚(浙江大学计算机辅助设计与图形学国家重点实验室,杭州 310058;英特尔亚太研发有限公司,上海 200241)

摘 要
提出一种新的基于高斯过程动态模型的节奏转移方法。该方法能够准确、有效地将现有运动中的节奏信息转移到新的运动中去,适用于各种不同类型的运动。首先,使用短时(short term)PCA计算源运动的节奏点,组合3种重要的运动特征求解目标运动的特征点;然后,使用动态规划算法来找到两者之间的最佳匹配,最大化减少计算时间及对目标运动的修改;最后,使用高斯过程动态模型对目标运动进行学习,并在隐空间进行节奏化插值,最终合成新的节奏化运动。
关键词
Rhythmical motion synthesis based on Gaussian process dynamical model

Lv Pei,Zhang Mingmin,Xu Mingliang,Li Ling,Song Qingjian,Pan Zhigeng(State Key Lab of CAD&CG Zhejiang University;Intel Asia-Pacific Research & Development Ltd)

Abstract
In this paper, we provide a new rhythm translation method based on Gaussian process dynamical model which can translate existing rhythm information to various types of motion accurately and effectively. Firstly, Short-Term PCA is used to calculate rhythm points in source rhythmical motion, three important motion features are combined to extract the motion feature points in target motion; Secondly, dynamical programming is utilized to find the best match between them in order to reduce the estimation time and damage to the target motion; Lastly, the Gaussian process dynamical model is applied to learn the target motion and rhythmical interpolation is done in latent space to get the new final rhythmical motion.
Keywords
QQ在线


关注微信