The Research Of Three-dimensional Motion Retrieval Based On SOM Feature

A large number of three-dimensional human motion capture data have been generated because of progressive development of motion capture technology and equipment to technical has been promoted.It have also been applied to a variety of areas ,such as computer animation, film special effects and other areas. Therefore, the research on motion capture data has become focus on the field of computer vision, graphics and animation applications.However, it need to start with retrieving the required movement sequence from the existing motion capture database quickly and accurately, before making the necessary changes about processing and synthesis, etc., to the motion capture data. Therefore, how to use computer effective technology to retrieve the needs of the various sports automatically ,quickly and accurately from the capture database is a serious problem.This article propose two kinds of algorithms about 3D motion retrieval measure based on SOM Feature and PCA index as well as based on SOM feature and weighted mahalanobis distance,they all based on the follows process of use of the topological characteristics of self-organizing maps (SOM) for feature mapping to human motion capture data.To simplify the process of motion retrieving, we use the characteristics mapping of SOM to achieve a combination of feature extraction and data dimension.While the topology of general SOM must be handled in order to enhance the feature extraction. After using SOM to doing feature mapping about each movement to the surface, on kind of thinking is using principal component analysis (PCA) feature extraction algorithm to extract the largest surface feature vector ,and construct indexing mechanism to speed up retrieval rate; one idea is using PCA to extract principal components based on the characteristic surface, and then using the contribution of principal components to determine the weight of the weighted Mahalanobis distance,At last calculating the weighted Mahalanobis distance for the similarity comparison. In this paper, two algorithms were simulated and compared, the experimental results have been proved that the two methods are effective.