摘要
为了解决由于运动捕捉数据的高维性导致传统低级算法不能很好分割动作的问题,提出了一种基于相对位置的2阶段低级分割方法.该方法使用末端关节相对于根关节(臀)的三维位置信息作为特征,首先采用巴特沃斯滤波器对选取的特征进行降噪,之后结合过零点检测和阈值方法找出各维分割点,并将属于同一个末端关节的三维数据上的分割点进行综合,从而获取各个末端关节的分割点,最后从各个末端关节的分割点融合出全身分割点.从卡耐基梅隆大学运动捕捉数据库中选取不同种类动作共12 941帧用以评价所提出的方法.结果表明,该方法能够有效地对动作捕捉数据进行分割,同基于速率的2种方法相比,所提出方法可以获得更高的查全率与准确率.
To solve the problem that traditional low-level segmentation methods can hardly achieve suitable segmentation for motion data due to high dimensionality,a two-phase low-level motion data segmentation method was proposed based on relative position. The positions of end joints relative to root joint were utilized as the features and were later processed to remove noise by Butterworth filter.According to zero-crossing and thresholding methods,the segment points for each dimension were located in three dimensions of end joint and synthesized to get segment points for end joint. Those segment points of all end joints were merged to derive the segment points for whole body. The 12 941 frames of motions from CMU Mocap database were explored to evaluate the proposed method. The results show that the proposed method is effective for segmenting motion capture data. Compared with the 2 velocity-based segmentation methods,the proposed method can lead to better recall and precision ratio.
作者
杨洋
詹永照
王新宇
YANG Yang ZHAN Yongzhao WANG Xinyu(School of Computer Science and Communication Engineering, Jiangsu University, Zhenjiang, Jiangsu 212013, China)
出处
《江苏大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2017年第2期186-191,共6页
Journal of Jiangsu University：Natural Science Edition
基金
国家自然科学基金资助项目(61402205)
中国博士后基金资助项目(2015M571688)
江苏大学高级人才启动经费资助项目(13JDG085)
关键词
动作分割
运动捕捉
高维性
过零点检测
相对位置
segmentation
motion capture
high dimensionality
zero-crossing detection
relative position
