摘要
针对人体运动跟踪的特点 ,在退火粒子滤波方法的基础上 ,提出基于序列蒙特卡罗方法的 3D人体跟踪算法 通过状态空间分解提高了退火系数选择的鲁棒性 ;同时 ,在每次退火时采用PERM采样方法 ,而不是标准的重采样 ,能在一定程度上抑制观测模型与真实分布之间的误差 ,从而提高算法的稳定性 通过模拟实验表明 。
Characterized by its high dimension and multi-modal, the 3D human motion tracking has been a stubborn problem in computer vision for many years. This paper introduces a new sequential Monte Carlo (SMC) method, based on annealed particle filtering, to solve the problem. Firstly, the new method adopts the state space decomposition in conjunction with simulated annealing to improve the annealing efficiency in a comparably lower dimension; next, the PERM sampling after every annealing, instead of standard resampling, is used to compensate the error between the observation model and the true target distribution. At the end of this paper, a simulated experiment is given to show that our improved SMC-based method is capable of tracking 3D articulated human.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2005年第1期85-92,共8页
Journal of Computer-Aided Design & Computer Graphics
基金
国家重点基础研究发展规划项目(2004CB318000
G1998030608)
国家"八六三"高技术研究发展计划项目(2001AA231031)
国家科技攻关计划课题奥运科技专项(2001BA904B08)
关键词
跟踪
随机采样
序列蒙特卡罗
模拟退火
PERM
tracking
random sampling
sequential Monte Carlo
simulated annealing
PERM
