摘要
针对人体运动捕捉数据缺失问题,提出一种结合模糊聚类和投影近似点算法的缺失数据重构恢复方法.首先对不完整运动序列矩阵的缺失数据位置进行线性插值预处理,粗略补全矩阵以得到较完整的运动序列;然后利用模糊C-均值算法将粗略恢复后的复杂人体运动数据细分为含有多个不同语义运动片段的时序组合;再根据相同运动语义片段数据矩阵存在低秩特性,对细分后相应的各原始运动子片段采取投影近似点算法进行缺失数据恢复,并按照运动片段的时序特性进行组合;最后将原有未缺失数据与其相应位置重构恢复后的数据进行置换,根据人体运动轨迹的局部线性特性进行线性平滑,以保证运动序列的连贯性,从而达到对整体运动捕捉数据重构恢复目的.实验结果表明,该方法能够有效地对缺失运动数据进行恢复,使得重构后的运动序列能够较好地逼近于真实运动轨迹,准确度较高.
It is a challenging topic to recover the missing values within the motion capture data. To address this issue, this paper presents an effective missing human motion capture data recovery approach via fuzzy clustering and projected proximal point algorithm（Pro PPA）. The proposed approach first utilizes the linear interpolation method to roughly fill the missing values within the incomplete motion matrix. Then, the fuzzy c-means clustering is employed to separate the previously processed motion sequence into several motion sub-clips, and each sub-clip motion incorporates a particular semantic behavior. Subsequently, according to the low-rank property within each sub-clip motion matrix, the Pro PPA algorithm is selected to achieve matrix completion and the missing values with respected to the original motion capture data can be recovered. Finally, the recovered positions of the non-missing values of the temporally combined sub-clip motions are further replaced by the corresponding original data and linearly smoothed on the basis of the local linear property of human movements simultaneously, whereby the whole incomplete motion sequence can be well reconstructed. The experimental results have shown that the proposed approach is able to well perform the incomplete motion recovery with high accuracy, and the recovered motions can well approximate the real moving trajectories with satisfactory performance.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2015年第8期1416-1425,共10页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(61202298
61202299
61300138)
福建省自然科学基金(2014J01239
2015J01656)
华侨大学高层次人才科研启动基金(14BS207)
关键词
缺失数据重构
投影近似点算法
模糊C-均值
线性插值
矩阵补全
missing values reconstruction
projected proximal point algorithm
fuzzy c-means
linear interpolation
matrix completion
