基于关键帧的三维人体运动检索

3D Human Motion Retrieval Based on Key-Frames

提出一种基于关键帧的三维人体运动检索技术.首先采用骨骼夹角作为对原始运动数据的特征表示并提取运动片断关键帧集合,以此作为原始运动片断的特征表示;然后基于所提取关键帧数据在相似运动片断之间具有一致性的特点,在两两关键帧集合之间建立距离矩阵进行相似度匹配.实验结果表明,与现有大多数基于内容的三维人体运动数据检索方法相比,该方法能够取得更好的时间效率且不依赖于任何预先设定的参数,相对于现有基于关键帧的运动检索技术而言,其能够获得更好的检索效果.

A content-based motion retrieval approach based on key-frames is proposed. For each motion clip in motion database, the bone angles are employed as motion features and key-frames can be selected based on them. When retrieval command is issued, the distance matrix is constructed between key-frame set of query example and that of a motion clip from database. Then the similarity between them is calculated based on this distance matrix. To validate this retrieval approach, we test it on a database containing about four hundreds motion clips which can he divided into various kinds of human movements. The experimental results demonstrate the effectiveness of our approach.