摘要
针对传统的低级时域分割算法不能很好地应用于运动捕捉数据(具有高维性)的问题,提出了一种基于加权置信度的运动捕捉数据低级时域分割算法.首先计算每个时间点在各维上作为分割点的置信度;之后将这些置信度按照对应的过零强度进行综合,得出加权置信度;最后通过寻找加权置信度的局部最大值点,并根据合适的阈值加以限定,确定出全局的分割点.通过测试多组数据,选择得到最优结果时的阈值.用卡耐基梅隆大学运动捕捉数据库的数据进行了试验.结果表明,该方法可以有效地对动作捕捉数据进行分割;在最优阈值条件下,该方法在各方面指标上均显著优于基于曲率或全身速率的方法.
To solve the problem that traditional low-level temporal segmentation methods could hardly work out a suitable segmentation for motion capture data with high dimensionality,a low-level temporal segmentation algorithm was proposed based on weighted confidence. The confidence as segment boundary of each time point was calculated in each dimension,and the confidence values of different dimensions were incorporated by zero-crossing strength. The whole body segment boundaries were derived by pruning the local maximum and determined by an optimum empirical threshold. The optimum threshold was obtained by testing several sets of data. The experiments were completed based on the data from motion capture database of Carnegie Mellon University. The results show that the proposed method is effective for segment motion capture data,and under the optimum threshold condition,the method is far better than the method based on curvature or whole body rate.
出处
《江苏大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2015年第3期312-318,共7页
Journal of Jiangsu University：Natural Science Edition
基金
国家自然科学基金资助项目(61402205)
高等学校博士学科点专项科研基金资助项目(20113227110021)
江苏大学高级人才启动经费资助项目(1281170027)
关键词
运动捕捉
低级时域分割
高维时序数据
过零点检测
加权置信度
motion capture low-level temporal segmentation high-dimensional time series zero-crossing detection weighted confidence
