一种人体运动相似性度量方法

摘要

This paper uses continuous geometric features to present motion sequence and obtain their subspace respectively.Due to the difference of complexity between different categories of motions, their intrinsic dimensions are not the same.It introduces a method which is able to measure the distance between subspaces with different dimensions.Experimental results demonstrate that improved continuous geometric features are efficient and it outperforms than supervised method, in addition, its response time and storage space are quite low.%使用连读几何特征表示运动序列,分别得到各自运动序列的子空间表示,由于不同类别运动的复杂性差异较大,其内蕴维度也不同,因此引入一种可以度量不同维子空间距离的方法计算运动相似性.实验结果表明,与监督方法相比,该方法能够获得更高的查询精度,且没有任何人工干预,可应用于自动检索.另外,其查询时间较少,存储空间较低.