Human Motion Retrieval Based On SOM And SAX

In recent years, the motion capture data is widely used in computer animation. While traditional motion capture data retrieval methods based on numerical similarity have high computational complexity, such as the keyframe-based method, dynamic time warping method and etc, thus, a novel retrieval method based on logic similarity is proposed in this paper. The basic idea of the proposed method is to extract the motion motif of the entire database, which is a type of the common characteristics of the movement used to reduce the numerical similarity metric impact on retrieval algorithm,At first, the fundamental theory of motion capture data is illustrated in detail in this paper, which describes the basic framework of obtaining motion capture data, in the following the motion similarity is discussed include numeric similarity and logic similarity, and further more, the aspects influencing motion similarity and the similarity metric are analysized; in the next, the proposed method is described in detail, firstly the motion capture data is converted to the quaternion representation, and then the self-organizing mapping algorithm is used to find the best matching unit with the clustering property of the self-organizing mapping algorithm applied to reduce the dimensionality of motion capture data, and then the symbolic aggregate approximation algorithm is used to symbolize the obtained data,by which the index mechanism is established, and further more, the multi-resolution motif extraction algorithm is exploited to obtain the motion motif.Based on the proposed algorithm, the retrieval system is established and the deficiency of the proposed alogrthm is found through the analysis of the experiment results, as for this, the proposed method is analysized and the improvement solution is offered, where the aligned cluster analysis for temporal segmentation algorithm is supplemented to segment motion, based on which the experiments are implemented further and the results indicate that the proposed algorithm can effectively retrieve similar motions.