Analysis and Comparison of Dimensional Reduction Based on Capture Data

Owing to the high dimension characteristic of motion in catching original data, the high dimensional original data will be projected into low dimensional sub space. The internal structure of body motion will be revealed through this low dimensional space. The elimination of the related redundant information of high dimensional characteristics becomes key technology for 3D motion capture data. This paper applies key frame and dimension reduction method based on several machine learning methods to handle motion capture data. After a series of experimental results, non-linear sub space is of better performance and wider availability.