Motion Segment based on Sparse Representation and 3D Features

With the emergence of a large number of 3D human motion capture database, which makes how to efficiently analyze and process the data of human body movement, and make use of the motion capture database become a new challenge. In order to reduce the high dimensional complexity of the data, firstly, a 3D dimensional feature based on 3D spatial and temporal characteristics is extracted from the motion of the human body, then, the motion data is re-expressed by the use of method for sparse representation, and different motion types are separated from long motion sequences, so that a motion database used for subsequent motion recognition and retrieval can be established. Introduction Currently, analysis of 3D human motion data still lacks complete and effective analysis and processing technology, which can not efficiently, rapidly, automatically and intelligently apply the large scale 3D human motion database to digital media field. The 3D human motion data contains abundant semantic meanings of objects, events, behaviors, and scenes, and its characteristics of mass, non structure, high dimension and multi order bring a great challenge to the semantic understanding. In recent years, compressed sensing and variable selection (when in the analysis of high dimensional data such as images, the variable selection is also called as feature selection in this application declaration) theories have been combined with methods, which are used for the formation of a more effective “sparse representation” on media data, becoming a hot issue in the fields of computer vision and machine learning, etc.. Compressed sensing makes the utilization of the future knowledge of “data being sparse and compressible” to achieve signal reconstruction, and in this aspect, some representative research works have been carried out by Donoho David and Emmanuel Candes of Stanford University, and Terence Tao (Tao Zhexuan) of University of California at Los Angeles, involving random matrix, signal recovery, sparsity measurement, etc. [1,2] In view of the superiority of compressed sensing and variable selection in data processing, Wright and Ma Yi of the University of Illinois, Urbana-Champaign have introduced it into face recognition, and a new thought of carrying out feature selection by using the l1-paradigm constraints model for recognizing human faces has been put forward[3]. Many features can be extracted from the media data, therefore, how to select effective sparse representation from high dimensional feature, and then study the theory and method of semantic understanding of media data on the basis of sparse expression has become a developing trend in the field of computer vision and pattern recognition, and it has been used in the visual word selection[4], image annotation[5], and image restoration[6] in succession. In the process of the identification of the real world, Urbana-Champaign have cooperated together to apply the sparse representation to visual object recognition, which has won the first prize in the PASCAL visual object recognition challenge (VOC2009) [7]. In order to realize this goal, aiming at 3D human motion data, we firstly make extraction of three-dimensional space-time characteristics, and then, the sparse representation of motion is given, and segmented. International Conference on Manufacturing Engineering and Intelligent Materials (ICMEIM 2017) Copyright © 2017, the Authors. Published by Atlantis Press. This is an open access article under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/). Advances in Engineering, volume 100