|
Increasing surge of multimedia data for fast data retrieval and query brought greater difficulties. However, the majority of users want to be able to quickly extracted from multimedia data the interested and implicit knowledge (concepts, rules, laws, patterns, and relationships), in order to use them for quick retrieval and query, while providing problem solving levels of decision-making support. This contradiction to the multimedia data mining challenges. Sports video moving target as the object of study, from the theoretical level and technical framework for system, from low-level motion feature to the layer-by-layer between the semantics of high-level sports video mining, not only to provide support for the user to find information quickly, and can provide decision support for the user to solve the problem. Theory and system framework, which first study sports video sports mining theory is divided into three levels: basic feature layer, mode, event layer, and knowledge layer. Moving mining method based on this on all levels of the exploratory study, the focus of a basic movement characterized in mining method, and a motion-based modes, events, and knowledge (Model, Event and Knowledge, MEK) mining methods. Thesis focused on moving target relatively small, relatively open space across the network sparring game, such as the tennis video badminton video and other types of video sports mining method. Mining method for the basic movement characterized the Camshift algorithm in the accurate extraction of a moving object on the basis of using the improved to solve the problem of tracking lost, for the extraction of subsequent trajectory, position and direction of movement of the moving object obtained foundation. MEK for motion-based mining, paper is mainly methodological research to a specific instance of the video. The main research contents and innovations are as follows: Sports the video sports mining theoretical level and technical framework. The theory main level contains three parts: the basic characteristics of layer mode, the event layer, and knowledge layer. In this theoretical level, respectively, corresponding to the technology roadmap: basic motion feature extraction techniques, mode, event mining technology. Theoretical level layers are not independent, low-level services for senior, but they also have their respective characteristics exists, separate for mining information. A basic motion feature mining methods. Basic motion feature mining framework on the basis of the analysis of sports video motion feature and a brief introduction of the various parts of this framework, and the corresponding basic feature extraction techniques. Wherein the extraction of the basic characteristics including: the locus of extraction, the extraction of the position of the moving object and the direction of movement of extraction, etc.. A motion-based mode, events, knowledge (MEK) mining methods. Mining is the use of the geometric characteristics of the correction trace motion-based model to analyze, and obtain some initial video skip the structured information, and based on statistics of exercise habits problems. Movement-based event detection includes two aspects: SEIT construction, matching based on SEIT events. Motion-based knowledge mining is the use of the traditional data mining methods, its mode to extract a more advanced knowledge. Designed and implemented a sports video sports mining platform - SVMMP (Sports Video Motion Mining Platform) mining theoretical level and technical framework for the application and validation of sports video motion. In summary, this paper presents the basic concepts of sports video sports mining technology framework and theoretical level, on the basis of the classical methods related processing technology has been improved, and sports video sports mining platform through the design and implementation, and validation of this article ideas. These studies video motion mining provides a new solution, the video moving continuous development and improvement of mining technology will make it to put decision-making in areas such as play an increasingly large role in the management and sharing of information resources and problem.
|