Research On Motion Capture Data Retrieval And Segmentation Technology

Enabled by the advances in motion capture (MoCap) technology, MoCap data are being increasingly generated and used in many fields including film production, video game development, interactive virtual reality, computer-assisted medical diagnosis and so forth. The cost of motion capture is very high, because the MoCap equipment is expensive, professional actors often need to be employed and the capturing process is usually time-consuming.In order to efficiently review or reutilize the existent data, many computer animation researchers have been focusing on how to analyze,retrieve and segment the MoCap data.Most existent algorithms for content-based MoCap data retrieval work only on full human body motion data. And they assume that the motion data are obtained for subjects with the same skeletal structure and joint labelling. However, different MoCap systems are being used in the world and, even with the same system, different schemes of marker attachment, skeletal structures and/or joint labellings could be adopted. So in this thesis, we propose an original and generic algorithm for content-based retrieval of heterogeneous MoCap data, which are obtained from different sources, with different marker attachment schemes, with different morphological structures, with different joint or marker labelling schemes, and/or for hybrid types of subjects.To facilitate the capture, raw motion capture data is usually a long motion sequence with several motion types, which is not convenient for data storage, retrieval and reuse. To this end, we put forward two algorithms for automatic MoCap data segmentation.The contributions of the thesis mainly include:1. Generic content-based motion capture data retrievalWe propose and construct a novel form of descriptor, which we call motion signature, to statistically describe both the high-level and the low-level morphological and kinematic characteristics of a MoCap sequence. Then the content-based retrieval is conducted by computing and ordering the motion signature weighted distance between the query and every item in the database.For maximized retrieval performance, we innovatively propose to pre-learn an optimal set of weights for each type of motion in the database through biased discriminant analysis, and adaptively choose a good set of weights for any given query at the run time. Excellence of the proposed scheme is experimentally demonstrated on various data sets and performance metrics.2. Motion capture data segmentation based on graph partitionFor better reuse of motion capture data, long motion sequences need to be segmented into multiple motion clips of simple motion types.We propose a method for motion capture data segmentation based on graph partition. Each frame of a motion sequence is viewed as a node in an undirected weighted graph, and the weight of an edge is the similarity between two frames corresponding to the two nodes connected by the edge. The optimal segmentation is obtained through graph partition algorithm, which leads to high similarities of nodes in each subgraph, and low similarities between nodes of different subgraphs. After the segment scores at each frame are calculated, double thresholds decision method is conducted on the score curve to detect segment points. Experimental results show that our method obtains good segmentation results.3. Motion capture data segmentation by genetic algorithmTo sovle the problem of MoCap data storage and reuse, we propose a novel genetic algorithm approach to MoCap data segmentation. For a given MoCap sequence, it constructs a symbolic representation through unsupervised sparse learning, identifies some simple motion patterns by string analysis method, correspondingly detects the candidate segmenting points, models the selection/deselection of each candidate with a gene, and employs the genetic algorithm to find the optimal solution. In the experiments, our algorithm achieved good result both in segmentation accuracy and execution efficiency compared with representive segmentation algorithms in recent years.