Research On Keyframe Extraction And Retrieval Algorithm For Motion Capture Data

With the rapid development of computer animation, motion capture especially the human motion capture technology is also rising. It has important application in the fields like film and television entertainment, science and education and even industrial and military. However, the development of the technology also brings a lot of technical difficulties, including not only the performance improvement of the motion capture technology itself, but also the motion capture data processing operation such as editing, segment and merge, compression and reconstruction, retrieval, etc. And the data processing can greatly improve the reusability of motion capture data, which is the research focus in this field in recent years.This paper focuses on the two common data processing algorithms for human body motion capture data: keyframe extraction and retrieval. We propose an improved optimal reconstruction error keyframe extraction algorithm and an efficient retrieval algorithm based on index space. The main research work and content in this paper are as follows:1. This paper expounds the various coordinate systems including the cartesian coordinate system, which is used in motion data computing in three dimensional space. And the three kinds of angular dispalcement expression: rotation matrix, euler angles and quaternions, and the common computing and converting.2. This paper research the BVH format of human motion capture data, and explains the data definition of each part of the data and the corresponding human joint definition. At last the computing procedure of how to parse the data to the motion in three-dimensional space.3. This paper proposes an improved method of keyframe extraction based on optimal reconstruction error. The reconstruction error that can measure the importance of frames is defined firstly. Then we extract keyframes frame by frame and calculate the corresponding reconstruction error, and build up a reconstruction error curve finally. The optimal compression ratio can be determined according to the curve and the keyframe sequence can be extracted at the same time. Through contrast experiment, the keyframes extracted by the method of this paper show more representative and have good motion generality.4. This paper proposes an efficient motion retrieval method based on index space. Firstly, the dimensionality of the original motion data is reduced by key frame extraction, and then the original angles are converted into feature sequences based on the features which meet the human movement semantics. According to the feature sequences that are corresponding to all of the motions in database, we construct an index space based on pose-features. Retrieval is processed by matching pose-features sequentially on index space and it can find the matching motions in temporal consistency. The experimental results demonstrate the advantages of our approach and the method can be used to meet the needs of a variety of motion retrieval.This paper mainly proposes new methods aim at the keyframe extraction and the retrieval for human motion capture data. And the advantages of the methods are verified through experiments. At last, we present the expectation for the improvement of human motion capture technology.