CASIA OpenIR  > 博士后  > 出站报告
Keyword运动捕捉数据 关键帧提取 运动数据检索
Other Abstract
The data driven method synthesizes human animation by driving the avatar with
motion capture data, which has the advantages of high efficiency and high fidelity. So
this method has been widely used in many areas in recent years. However, due to the
expensive price of motion capture equipment, motion capture data reuse technology is
widely used to reduce the cost of data acquisition. In order to improve the efficiency
of data reuse, this report focuses on two key technologies: motion capture data key
frame extraction and motion data retrieval. The main contributions of this report are
listed as follows.
(1) A optimized key frame extraction method based on multi-scale pose saliency
is proposed. This method selects some motion components according to their variance
as the feature representation of human motion. The multi-scale saliencies of all
feature components are calculated and the posture saliency of the motion sequence is
evaluated from the weight mean of them. The initial key frames are extracted from
motion capture data based on postural saliency. Then the optimal ratio of compression
rate to reconstruction error is used to determine the optimal number of key frames.
Finally the corresponding number of key frames are extracted from the candidate key
frames by the optimized reconstruction error algorithm. Experimental results show
that this method can effectively extract the optimized key frames from motion capture
(2) A motion data retrieval method based on dictionary learning and spectral
clustering is proposed. This method is divided into two stages: motion semantic
clustering and motion data retrieval. In the First stage, The key frames are clustered
by K-Means Algorithm, and the similarity between key frames is calculated according
to cluster to generate a similarity matrix. Then spectral clustering algorithm is used to
cluster motion data into some semantic classes. Dictionary learning algorithm is used
to extract dictionary atoms for each semantic class. In the second stage, The key
frames of query sample are clustered by K-Means Algorithm, and one key frame is
selected from each cluster as the initial dictionary for dictionary learning. Then
candidate motion clips are retrieved from the motion library by the matching degree between dictionaries. DTW algorithm is used to calculate the similarity between
query sample and candidate motions. The retrieval results are presented in the order of
similarity. Experimental results show that the method proposed in this report can
efficiently retrieve the target motion.
Abstract数据驱动的人体动画合成方法利用预先捕捉的人体运动数据驱动虚拟角色 形成人体动画,具有动画合成效率高,所得结果真实感强的优点,近年来得到了 广泛应用。 但由于运动捕捉设备价格昂贵, 人们普遍采用运动捕捉数据重用技术 来降低数据采集成本。 为了提高数据重用效率, 本报告研究了运动捕捉数据关键 帧提取和运动数据检索两项关键技术,并取得了以下成果: ( 1) 提出了一种基于多尺度姿势显著性的运动捕捉数据最优化关键帧提取 方法。该方法首先根据方差筛选出部分运动分量作为人体运动的特征表示, 计算 特征分量的多尺度显著性并加权拟合形成运动序列的姿势显著性, 基于姿势显著 性提取运动捕捉数据的初始关键帧;然后基于初始关键帧计算最优化压缩率与重 建误差比,并由此确定最优化关键帧数量;最后利用重建误差分段最优化算法从 候选关键帧中提取出相应数量的最优化关键帧。实验结果表明, 本报告提出的方 法能够有效地实现运动捕捉数据的最优化关键帧提取。 ( 2) 提出了一种基于字典学习与谱聚类的运动数据检索方法。 该方法分为 运动语义聚类与运动数据检索两个阶段。在第一阶段, 首先对运动数据关键帧进 行 K 均值聚类, 根据聚类计算关键帧间的相似度, 构造相似度矩阵,然后采用谱 聚类算法对运动数据进行语义聚类,采用字典学习算法提取语义类的字典原子用 作运动数据的特征表示。在第二阶段,首先对查询样例的关键帧进行 K 均值聚类, 从每个类中选取一帧作为初始字典进行字典学习, 然后通过字典匹配从运动库中 检索出候选运动,用 DTW 算法计算查询样例与候选运动的相似度,按相似度顺序 给出检索结果。 实验结果表明本报告提出的方法能够实现相似运动的高效检索。
Document Type研究报告
Recommended Citation
GB/T 7714
刘云根. 运动捕捉数据重用关键技术研究. 2018.
Files in This Item:
File Name/Size DocType Version Access License
04_博士后出站报告_提交版.pdf(2573KB)研究报告 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[刘云根]'s Articles
Baidu academic
Similar articles in Baidu academic
[刘云根]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[刘云根]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.