基于低维流形的人体行为跟踪方法  

Human motion tracking method based on low-dimensional manifold

作　　者：何凯霖[1] 丁晓峰[1] HE Kai-lin DING Xiao-feng(Department of Computer Science, Chengdu College, University of Electronic Science and Technology, Chengdu 611731, China)

机构地区：[1]电子科技大学成都学院计算机系,四川成都611731

出　　处：《计算机工程与设计》2017年第5期1361-1365,共5页Computer Engineering and Design

摘　　要：为保证人体行为跟踪的准确率,降低算法的复杂度,提出一种基于低维流形的人体行为跟踪方法。采用四元数描述人体姿态,防止欧拉角的万向锁问题,通过低维流形降低粒子滤波识别与跟踪所需的粒子数量,降低算法的复杂度,保证粒子滤波带来的高性能。实验结果表明,该方法拥有比经典对比算法更好的准确度和更低的开销。To reduce the complexity of human motion tracking, a low-dimensional manifold based particle filter method was pro-posed Quaternions were used rather than Euler angles to avoid gimble lock problem. Low-dimensional manifold was used to re-strict the scale of particles for reducing complexity and achieving high accuracy. Experimental results show that the presented method has higher accuracy and lower cost than state-of-the-art control methods.

关 键 词：低维流形 人体行为跟踪 粒子滤波 姿态估计 能量消耗 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]