基于运动能量模型的人体运动捕捉数据库的检索

Retrieval of Human Motion Data Based on Energy Model

不同于经典的基于几何空间位置的检索方法,首先提出基于运动能量的描述人体运动的模型,在此基础上引入熵的概念,提取能体现运动特征的关键关节作为衡量动作相似性的基准;然后利用Keogh定界算法建立索引,以加快检索速度;最后,利用动态时间变形算法计算运动例子和检索集之间的相似度,确定检索结果集.实验结果表明,该方法速度快、准确性高.

A novel method based on kinetic energy is proposed for the retrieval of human motion, which is different from the traditional retrieval method based on spatial position. The entropy is used to extract key joints, which are taken in the later steps to compare different motions. To speed up retrieval, the index is built up based on Keogh lower bound. Finally, the dynamic time warping distance is used to measure the similarity of different motions. Experimental results show that our method is accurate and efficient.