摘要： 人体运动捕获技术的发展使得运动捕获数据不断积累,人体运动的检索技术成为运动数据管理和重用过程中的关键环节。由于逻辑相似的运动在数值上并不一定相似,使用欧式距离度量2 个运动间的逻辑相似性难以取得理想的结果。为此,提出一种半监督的距离度量学习算法,利用带标记的运动和未标记运动进行训练以得到运动间的马氏距离度量,从而判断2 个运动之间的逻辑相似性,实现运动检索。实验结果表明,与现有的大部分检索算法相比,该算法能够得到更高的查询精度,且没有任何人工干预,可应用于自动检索领域。

关键词: 人体运动捕获数据, 半监督, 距离学习, 相似性度量, 检索, 识别精度

Abstract: With the rapid development of the human motion capture technology,large amount of motion capture data is gradually accumulated,then human motion retrieval and recognition technology becomes the essential issue for motion data management and reuse. Logically similar motions may be numerically dissimilar,so it is difficult to get feasible results if the logical similarity between two movements is measured with Euclidean distance. This paper presents a semisupervised distance learning method for measuring the logical similarity with Mahalanobis distance which is trained by labeled and unlabeled motion data. Experimental evaluation result of the method shows that the proposed method is effective for motion retrieval.

Key words: human motion capture data, semi-supervised, distance learning, similarity measurement, retrieval, identification precision

中图分类号: 

• TP391