Research On Motion Mining Technique For Sports Video Analysis

Multimedia data is acutely increasing, which makes fast index and query more difficulty. However, most users expect that interested content and hidden knowledge (conception, regulation, rule, mode, relationship, etc.) can be quickly extracted from multimedia data, in order to make index and query faster and to support decision-making for problem solving levels. This conflict makes multimedia data mining meet huge challenge. This paper regards motion objects in sports video as research objects. We study from theory level to technique frame,and achieve step by step mining among video low-level motion feature and high-level motion semantic. This work provides not only fast query, but also decision support for problem.Firstly, this paper studies sports video motion mining (SVMM) theory level and system frame, in which theory level contains three levels which are basic feature level, model and event level, and knowledge level. We study motion mining methods for each level, and emphatically propose a method of basic motion feature mining (BMFM) and a method of Model, event, and knowledge mining (MEKM). This paper mostly studies racket games which contain few motion objects and open field, such as tennis, badminton, and so on. For BMFM, we use improved Camshift algorithm to solve track-lose problem after accurately extracting motion object, which confirms base for track extraction, motion object position, and motion orientation. The main content and innovations are as follows:Propose SVMM theory level and corresponding technique frame. The theory level contains three parts: basic feature layer, model and event layer, and knowledge layer. Under the theory level, we propose technique route: basic motion feature extraction technique, model and event detection technique, and so on. Each layer in this theory level is not absolute. Low layer serve for high layer, and each layer has its characteristic, which could absolutely mine information.Study a BMFM method. After analyzing sports video basic motion feature, we propose BMFM basic frame, and briefly introduce each function in the frame and basic feature extraction technique. Basic feature extraction contains: tract extraction, position extraction, and orientation extraction.Study a MEKM method.Model mining based on motion uses the geometrical feature of modificatory track (MT) to analyze, and achieve rude structure information and motion habit problem based on statistics. Event detection based on motion contains two aspects: the creation of the SEIT and event matching method based on SEIT. Knowledge mining based on motion extracts more advanced knowledge by traditional data mining method and the result of model mining.Design and achieve the SVMMP, which apply and validate the SVMM theory level and corresponding technique frame.To summarize, this paper proposes SVMM basic concept, technique frame and theory level. We improve the classical disposal technique, design the SVMMP, which validate the technique route. These researches provide a new solution for sports motion mining problem. The development on video technique will bring into play important role in the field of information resource management and share and problem decision-making.