A novel automatic motion capture data recognition method based on statistics learning and subspace. (19th July 2010)

• Record Type: Journal Article
• Authors: Xiang, Jian; Zhu, Hongli
• Is Part Of: International journal of computer applications technology. Volume 38:Number 1-3(2010)
• Page Start: 49
• Page End: 54
• Publication Date: 2010-07-19
• Subjects: motion control -- RBF networks -- feature extraction -- reinforcement learning -- motion capture data -- motion recognition -- neural networks -- motion classification -- human joints; Technology -- Data processing -- Periodicals; 620.00285
• Languages: English
• ISSNs: 0952-8091

Abstract

In this paper, we propose a motion recognition method based on motion capture data. To recognise motion type, a generalised Isomap non-linear dimension reduction based on Radius Basis Function (RBF) networks and feature extraction is used to project original motion data into low-dimensional subspace. Then, some motion-type classifiers are learned for each human's joint in subspace. Then, we use ensemble reinforcement learning to enhance learning results. Experimental results show that our methods are effective for 3D human motion recognition and control.