Human Motion Capture Data Retrieval Based on Quaternion and EMD

Abstract

In this paper, a novel human motion captured data retrieval approach is presented Based on Quaternion and EMD. The method mainly contains two steps: indexing and matching. In indexing part, for solving high dimension data problem, we use the quaternion to represent key-joints rotation information, and mapping the distribution of original CMU database, we take K-means clustering to categorize query and candidate features in datasets. In matching part, according to the clustering results, the distance matrix of each feature dataset is established. The next, the EMD measure algorithm is employed to match between motions, and similarity scores are obtained. Experiment results show that the proposed approach is efficient, and it is superior to existed methods.