利用共享动态隐变量模型估计三维人体运动  

Shared latent dynamical structure for three-dimensional human pose estimation

作　　者：仝明磊[1] 韩红[2] 朱武[1] 

机构地区：[1]上海电力学院计算机与信息工程学院,上海201300 [2]西安电子科技大学电子工程学院智能信息研究所,西安710071

出　　处：《中国科学：信息科学》2012年第1期47-56,共10页Scientia Sinica(Informationis)

基　　金：国家自然科学基金(批准号:61075041);上海电力学院新进教师基金(批准号:K-2010-16);中央高校基本科研业务费专项(批准号:JY10000902042)资助项目

摘　　要：可视媒体中的基于学习的三维人体运动分析是计算机视觉领域中非常有挑战性的课题.本文在Gauss动态隐变量模型与共享隐结构的基础上,给出一种新的共享动态隐变量模型用于三维人体运动跟踪.该模型针对高维非线性动态系统,可以计算出高维状态向量和高维观测向量的共享动态低维隐变量,同时也能计算出隐变量对高维状态向量、高维观测向量的双向映射、以及隐变量自身的动态关系.使用该模型可以将传统的高维人体运动估计分层为先估计低维隐变量状态,再重建高维人体运动.在实验结果中,用仿真图像序列与真实图像序列证明了方法的有效性.In this paper, a shared latent dynamical model （SLDM） and its application in tracking 3D human motion from monocular videos are proposed by combining the ideas of Gaussian processes dynamical model with shared latent structure. When tracking in high-dimensional space, SLDM can map state space and observation space to a shared latent space of low dimensionality with associated dynamics. During off-line training, three mappings, including dynamical mapping in latent space and mappings from the latent space to both state space and observation space, are learned. This model can separate traditional human motion estimation in high- dimensional space into two steps： In the first step, the shared latent dynamical variables are estimated; in the second step, the human pose of high dimension is reconstructed. Experiments in human motion tracking from monocular videos using simulations and real images demonstrate that this human tracking method is efficient.

关 键 词：共享动态隐变量 三维运动估计 计算机视觉 粒子滤波器 方差反馈 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]