An Efficient Method for Motion Retrieval

Description

Motion capture animation has attracted a lot of attention in recent years, as it provides an efficient way of obtaining realistic motion to drive an animation character to move around in a virtual environment. However, as the motion database gets larger, retrieving a suitable motion file to drive a character to move in a desirable way during a real-time application has become a bottleneck. In this project, the researchers propose to develop an efficient method for motion retrieval. By modelling the matching problem of motion sequences as a transportation problem, they convert the matching problem into an optimization problem of computing the minimum amount of energy required to morph one motion to another. An efficient and accurate method to calculate such morphing is called Earth Mover's Distance (EMD). EMD has been used in matching multimedia data, such as images. However, to apply EMD to motion matching and retrieval, a few issues need to be addressed, including motion feature extraction, strayed matching, and motion indexing. In this project, the researchers will develop an EMD-based motion retrieval method and address the three issues mentioned. They will also evaluate the performance of their EMD-based method against existing methods.