The Study Of Human Motion Time Series Visualization And Multi-Index Retrieval

The Analysis of human motion time series has a wide application field for animation, bionic design, sports and health care. It has become an important area about data mining research. With the rapid development of camera and sensor technology, the human motion capture technology has been more and more sophisticated. Meanwhile, how to solve the inherent problems of human motion based on the mass of human motion capture database, the traditional method could not understand the deep knowledge of human motion.This thesis introduces the forward kinematics method based on the properties of human motion database, and processes the human motion capture data by rotation matrix and homogeneous transformation matrix to transfer the value of joint angles into space coordinates in Cartesian system.The thesis uses the manifold learning method to reduce the dimensionality of human motion time series and projects in the low-dimensional space to achieve the high-dimensional time series visualization to express the trend of human motion and looks for potential motion split points.Secondly, the paper symbolizes the human motion by geometric relations to divide those motions into multiple properties and proposes a multi-index structure according to those properties. Comparing with traditional index structure, this multi-index emphasizes the using of the strength of the relationship between attributes. Therefore these properties could be combined into different characteristic sequences and then establish a multi-index structure according to these sequences. In the retrieval process, this thesis firstly determine the property changes or not by variance ratio and find the set which to be retrieved. Then use the limited-LCSS algorithm to measure the similarity of time series subsequence and integrate the several search results of characteristic sequences to achieve the final outcomes.Experiments show that the method of dimension reduction based on manifold learning can visually reflect the human motion trends and the multi-index subsequence retrieval algorithm could ensure the precision rate of the experiments.