Research On Semantic Method For Human Motion Capture Data

Human movement is ubiquitous in people's production and life, and 3D human motion data is a typical time series data. As the development of time series data mining technology, in recent years,3D human motion data application technology became more and more widely used and has made enormous economic worth. The movie such as Avatar, Planet of the Apes Wars and other 3D action movies earned brilliant and record-breaking box office. While throughout the existing analysis model of human motion processing, most of the general researches are based on the characteristics of the raw data. Such methods have gained much achievements, but there are bottlenecks to them that after studying on mechanism another scholars pointed out that the sequences similar on logical should not necessarily similar on numerical value and existed problem which called the semantic gap. Furthermore, methods based on raw data numerical values are not deep considered the domain feature of the human motion data, characteristics. Thus, there are certain limitations on understanding the semantics of the motion data and related semantic processing.This paper mainly analyzed the major existing method based on raw data numerical values and the existing applications of semantic methods and model. And this paper researched the ideas of the semantics and the process of architecture of the semantic network on human motion, analyzed human motion data through the different semantic levels. Firstly, this paper analyzed the raw numerical feature of human motion data to derive set of the basic elements of movement, then on these basic elements to generalize the formal description of the concepts of body motion and its corresponding organization.In this paper proposed the underlying semantic annotation method for human movement. The method anticipated human joints relationship from skeletal features, taken ideas and critical points of the human body modular approach from sport physiology, ergonomic human engineering and persona drafting, and researched to build the semantic representation of human joints, then sorted out the semantic description of parts of the basic elements of human movement which called the sense object in accordance with the whole-part semantic attribution. At the mean time, the method extracted spatio-temporal features of raw motion capture data and quantified different semantic motion in the semantic category space to the corresponding sense object. After clustering, analyzing and semantic marking, this paper finally obtained the basic definition elements of the original action: sense operation. On the basis of these two basic elements sets, the method related different senses with the subject-operation semantic relations to establish scale stable sense library consisted of the basic domain knowledge description of human normal motion.On the studying the organization and mining model of human movement knowledge, this paper proposed semantic knowledge mining model for human motion capture data. The model based on the human cognitive habits, and deeply analyzed domain knowledge of human movement from athletic training, course of ball game training and aerobics training. Then summarized the semantic relations, such as hyponymy, space constraints, successor relationship and precursor relationship, which conceal among different semantic movement concept, and employed idea of the organization of concepts which from WordNet and HowNet semantics to construct and represent these motion semantic concepts and the relationships among them by XML format. The model in this paper helps to fuse semantic relevance between high-level human motion domain knowledge and the basic underlying knowledge elements. As a result of the network structure in the model this paper proposed, computer can be better contents the data user's demands of domain knowledge, and computer can provides considerable capacity on human motion domain knowledge mining.