Application And Research On Intelligent Processing Algorithms For Motion Capture Data

The technique of motion capture (Mocap) is an important approach for motion data acquisition, and it is a new rising technique in the intersection of multi-disciplinary such as computer animation, maths, computer graphics, optics and so on. As its powerful function and important application value, more and more researchers have focused their interest on motion capture. The motion capture technique not only has meaningful research value, and it is very useful also. Now this technique has been widely used in computer animation, movie making,3D game, medical analysis, physical training and so on.With the development of the motion capture, some problems encountered in the utilizing of motion data have been proposed. Many research works have been done for these problems by the researchers abroad and at home, and lots of fruitful achievements are obtained. But there are still some open problems which have not been resolved. Aims at the key problems in motion capture data processing and application, some relative work has been done in this thesis, the main work can be concluded as follows:1) Scattered data processing for passive optical motion capture data:in the passive optical motion capture system, there are three important research topics, they are noise data processing, scattered data matching and missing data fitting. Aim at these problems, an approach based on template matching is proposed for the processing of motion capture data. In noise data processing, a clustering method is used which could combine the noise data in a points cluster into one available point. In the scattered data matching, the human template is segmented to several rigid sub-templates by the local rigid structure. Matching point-set with least error is searched for each sub-template from the human motion capture data by affine transformation, and the scattered data is identified by point tracking. The ambiguity of identification is notability reduced. In the missing data processing, two different missing data fitting method are proposed corresponding to the different situation, if a sub-template contains more than four markers, the missing data can be fitted by the relationship of its neighbors; but for the knee joint point, there only two markers in a rigid structure, in this situation we use the local linear prediction method to fit the missing data. At last, in order to proof the validity of the data processing method, some experiments are carried out, and the experiment results are satisfactory. 2) Data reusage techniques for human motion capture:the feature representation of human motion capture data based on least-square (L-S) distance is proposed. Based on the feature representation a similar pose retrieval method and a motion sequence segmentation method are designed. The similarity measurement of different human motion poses is defined by the least-square distance, and a similar pose retrieval method is designed which can be used to search similar poses from huge motion data base quickly. Under the feature representation, the human motion sequence can be simplified into a one dimension feature curve, and the motion sequence is segment into some motion units. At last, we can segment the motion sequence into some semantic primitive motion clips by the motion units clustering and hierarchy structure analysis of human motion.3) Robot control based on motion capture technique:a semantic based motion retrieval method if proposed for the simulation of space robot which supports the on-orbit module replacement. First, the motion data of space robot is acquired by Mocap equipment, and the data base is constructed by the Mocap data. Then, the least-square distance of two3D points’ sets is used to define the similarity between two motion poses of the space robot, and a motion retrieval method is designed by the similarity measurement. At last, we reconstructed the Mocap data and use the reconstructed data to drive the model of space robot. In order to prove the efficiency of the method proposed in this paper, some experiments is carry out, the experiment results indicate that the motion sequences got by the retrieval method is smooth and consecutive and which demonstrate the efficiency of the method proposed in this paper.To sum up, this thesis presents the corresponding algorithms and implements methods surrounding about the main research content. The corresponding experiment platforms are constructed and the validity of the algorithms is examined. All the study will provide the theoretical basis and algorithmic rule for the design and development of software system.