Research On Some Key Technologies Of Motion Capture For Digitizing Kinetic Arts

Kinetic arts, included Chinese folk dance, opera themes and Chinese kung fu, are extremely rich in folk resources. Digital recording can prevent kinetic arts of traditional cultures lost with the death of the heir. Motion capture data is more significantly effective to record human motion in data expression, data reuse and data management compared with video recording and motion notation.In this thesis, we concentrate on the issues of motion capture data acquisition and motion capture data management. For eliminating portability and convenience of the marker-based motion capture system, we focus on the study of markerless motion capture. Markerless vision-based motion capture provides a non-invasive solution making it very attractive. We mainly studied foreground (person) extraction with unknown backgrounds and robust pose estimation under the generative framework. For motion capture data management, we concentrate on the study of interactive nature of motion capture data retrieval method. Main research work and contributions of this thesis include the following aspects:1. Proposed a cosegmentation method based on local spectral for foreground extraction. Extracting foreground objects from an unknown background image is a challenging work because of the variety and complexity of the object and the background. In this thesis, we proposed a new seeded image cosegmentation method based on a local spectral method which combines bottom-up information and seeds’ knowledge effectively for segmentation. The cosegmentation problem is referred to as segmenting the same or similar foreground objects simultaneously from a group of images. Multiple images are connected into a weighted undirected graph so the cosegmentation problem is converted into a graph partitioning problem that is solved by Biased Normalized Cuts. The method is suits for multi-view image foreground extraction with unknown background model and background complex scenarios, increases the flexibility of the motion capture system.2. Proposed a novel human pose estimation based on the topology consistency of closed point pairs. In general, the problem of pose estimation is an optimization problem under generative framework, but cost functions easily trapped in a local minimum based on extrinsic similarities (silhouette, edge, etc.) and pose estimation fail if the body is largely self-occluded. In this work, we define an effective cost function that combines extrinsic and intrinsic similarity, which can be translated by projecting a mesh model onto camera views to best match the silhouettes from an extrinsic point of view, and attempts to preserve the intrinsic geometry of the shape using predefined mesh models based on heat diffusion. This problem is a multi-dimensional non-linear optimization problem and effectively solved by stochastic search, e.g. Annealed Particle Filtering (APF), Interacting Simulated Annealing (ISA). However, the numbers of particles and layers are fixed in these methods. We present a dynamic method for annealing particle filter, which dynamic adjust the number of particle and layers based on intrinsic consistency of shapes. When self-occluded occurs or the reconstructed body shape topological changes, the intrinsic distance is larger and need more particles and layer numbers, others less. The results of experiment show that our proposed method is more robust for motion capture and cost a little time.3. Proposed a motion capture data retrieval method based on Labanotation, which is a 2-D graphical editable movement symbols. With the accumulation of motion capture data, quickly and efficiently find the desired movement data from the database is necessary. But not as accessible as images, direct input motion capture data as search samples is difficult. We use Labanotation to describe the desired motion as input, and also use Laban notation to index motion capture data and rank the similar motion data with the similarity based on Laban Movement Analysis (LMA).Combining above methods, we build a feasible human motion digital platform, which contains data acquisition, data storage, data management, and data display based on multiview video, motion capture data and Labanotation. Based on this, we protect kinetic arts of cultural resources completely.