基于DTW方法及Hermite插值的运动序列长度归一化

Normalization of Motion Sequences Based on DTW and Hermite Interpolation

运动捕获设备采集的数据冗余量大,缺乏结构化信息,编辑重用难度大,需用运动编辑方法处理。运动融合作为运动编辑的关键子问题之一,首要步骤就是运动序列长度归一化。提出一种基于动态时间规整方法(Dynamic Time Warping,DTW)及Hermite插值的运动序列长度归一化的方法,首先基于改进的DTW算法将运动序列进行姿态对齐,找到运动序列的一组对齐帧;其次按照长度的需要对运动序列对齐帧进行位移插值和角度插值,对根位移进行Hermite插值,对关节旋转角度采取四元数球面插值;最后,实验结果验证了该方法可以生成新的平滑的角色动画,提高了数据重用性,以此实现了运动捕捉数据处理效率的提高。

The motion capture data have a problem of large redundancy and lack structured information, thus lead to the difficulty of reusing. Motion editing is required to deal with this problem. As a sub- question of motion editing, the first step of motion blending is to normalize the length of motion sequences. An approach based on Dynamic Time Warping （DTW） and Hermite curve interpolation was proposed to normalize the length of motion sequences. Firstly, the motion sequences were warped by the improved DTW algorithm to get a set of aligned frames. Secondly, the root trajectory with Hermite curve and angle with quaternion spherical linear interpolation of the align frames were interpolated according to the required length. Experiments demonstrate that this method can generate smooth character animations, help to reuse the existin~ data, thus imDrove the t3rocessin~ efficiency of motion canture data.