Research Of Human Motion Capture Data Segmentation Based On Measured MDS

In recent years, with the rapid development of computer technology and deepening of the research in the field of computer graphics, 3D human animation has become an important research content in computer animation. Human motion capture technology through the record of the real actor’s body movement track, can obtain the real, exquisite human motion capture data. The efficiency and the level of 3D human animation production are improved effectively. The process of using human motion capture data often needs clear human motion clips with independent semantic information. So how to segment the complex and long human motion sequences with multiple semantic actions, and gain the motion segments with independent semantic information become an important issue that must be solved in human motion capture data processing algorithms.The main purpose of this thesis is to research the segmentation algorithm of motion capture data. The main works of this thesis are as follow:1. A segmentation algorithm of motion capture data based on measured MDS and improved oblique space distance is proposed in this thesis. The measured MDS was used in the proposed algorithm to achieve the space mapping from original high-dimensional data to low-dimensional, and the improved oblique space distance between frames in the specified windows and the preceding section in the low-dimensional space was calculated, and the final segmentation points was obtained by similarity detection. Finally the independent semantic motion clips were obtained, and the feasibility of the algorithm was verified by experiments, and the accuracy rate of our method is improved compared with the traditional algorithm.2. This thesis proposed a segmentation of human motion capture data based on information entropy feature selection and clustering. Firstly the information entropy of each dimension was calculated, and the data with more information was extracted, then the measured MDS was used to map the extracted data into low dimensional space, and the initial segmentation points can be got by the k-means clustering algorithm, finally the final accurate segmentation points were determined by similarity detection, and the independent semantic motion clips were obtained. The feasibility of the algorithm was verified by experiments. The accuracy rate of the proposed method is improved compared with other algorithms.Through the research, it can be found that when the motion capture data is mapped from high-dimensional space to low-dimensional space by the dimensionality reduction method used in this thesis, the correlativity and the original topology structure of the original motion capture data are better kept. Furthermore, the segmentation results got by the proposed motion segmentation methods are closer to the manual segmentation results, and the error rate is smaller, which verify the feasibility and effectiveness of the proposed methods.