Construction and application of a motion capture database using human motion characteristics

Abstract

Motion capture data is widely used to easily reproduce natural-looking human motions. However, it is very difficult to search a huge motion database for any particular human motion. A variety of motion retrieval techniques that enable users to find accurately corresponding motions have been proposed. We developed a construction and retrieval technique for the human motion database that considers the overall features of motion by using a dimension reduction technique and a machine learning algorithm. This technique enables efficient retrieval of a set of motions similar to the query. It is also possible to apply our technique to motion synthesis by creating a hierarchical motion graph based on the human motion database.