基于帧间距的运动关键帧提取  

Extraction of Motion Key-frame Based on Inter-frame Pitch

作　　者：李顺意[1] 侯进[1,2] 甘凌云[1] 

机构地区：[1]西南交通大学信息科学与技术学院,成都610031 [2]南京大学计算机软件新技术国家重点实验室,南京210093

出　　处：《计算机工程》2015年第2期242-247,共6页Computer Engineering

基　　金：国家自然科学基金资助面上项目(61371165);浙江大学CAD&CG国家重点实验室开放课题基金资助项目(A1416);计算机软件新技术国家重点实验室开放课题基金资助项目(KFKT2013B22);四川省动漫研究中心2012年度科研基金资助项目(DM201204)

摘　　要：运动捕获数据中有大量的冗余数据,不利于运动数据的压缩、存储、检索以及进一步重用。为此,提出一种基于帧间距的运动关键帧提取方法,提取代表运动捕获数据内容的关键姿势。利用四元数之间的距离表示人体姿态差异,将人体各个关节上的总变化作为帧间距,以运动数据首帧作为第一个关键帧。通过不断计算当前帧同最后一个关键帧之间的差异,消除差异小于阈值的帧,差异超过阈值的帧被当作为新关键帧。对提取关键帧集合采用四元数球面插值方法重构。为表现人体运动特征,在重构误差中引入关节速度分量,用人体姿势误差位置与人体关节运动速率之和表示原始运动与重建运动序列之间的重构误差。实验结果表明,该方法对原始运动既有较高的压缩率,又有较强的视觉概括效果。To solve the problem that the motion capture data has a large number of data redundancy,this paper proposes a key frame extraction method based on inter-frame pitch.In order to compress,storage,reconstruct and further reuse motion data,key-frame is needed to be extracted which represents the content of the motion.Quaternion is introduced to represent the difference between two rotations.The distance between two frames is defined by the total rotation differences and first frame is regarded as the first key frame.Then /calculate the difference between the current frame and the last key frame continuously.The frame is eliminated when the difference is smaller than the set threshold or the opposite is reserved for the new key frame.Spherical linear interpolation is used to reconstruct the sequence.To express the characteristics of human motion,the joint velocity is introduced.Reconstruction error is defined by the human body posture error of position and the motion speed error between the original frame and the reconstructed frame.Experimental result demonstrates that the original motion capture can be compressed in a high ratio and gives a good visual summary performance.

关 键 词：运动捕获 关键帧提取 帧间距 球面线性插值 重构误差 视觉概括力 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]