3D Human Motion Capture And Analysis For Mass Sports Teaching

With the rhythm of modern life become faster, people pay more attention to theirwork but ignore the importance of physical exercise. This is harmful to their health.The best way to solve this problem is to develop a Mass Sports Teaching systemwhich is cheap and easy to installation. Thus people can get better instruction whenthey want to learn a new set of motion. Further, such a Mass Sports Teachingsystem can help to improve the whole level of their physical fitness.A complete Mass Sports Teaching system involves3D human motion capture,human motion reconstruction and human motion analysis. After extensive productresearch and thesis review, we find that although there is such motion capture andanalysis system developed for the athlete, but it is so expensive that it can notaccepted by the masses. The Mass Sports Teaching system is still under researchand there are two problems as follows:1) Human models used in motionreconstruction are rough and regular, but the Mass Sports Teaching system prefersto lifelike and changeable models.2) The current research only focuses on motioncapture and reconstruction, but lack analysis of the problems occurs in motionreconstruction. According to the above problems we carry on such research asfollows.First, in the aspect of3D motion capture, to cut down the cost of the system, wechoose Microsoft Kinect as our motion capture equipment which has appropriateaccuracy and acceptable price. To track and record3D human motion effectively, weadopt a rendering pipeline. First, depth data is processed in the rendering pipelineprocess and matches with decision forest labeled data and generates the inferred bodysegments, the sensor then calculates the3D view from the top, front, and the left ofthe proposed joints. At last the sensor starts tracking the human skeleton and bodymovement based on the proposed joint points and the3D view.Second, in the aspect of3D motion reconstruction, to strengthen the effect ofmotion reconstruction and easy for visual comparison, we adopt MakeHumansoftware to generate lifelike models. To suppress the tremble of the motion and makethe plugin for the MakeHuman software compatible, we propose a method that reconstructing the motion based on Kinect SDK and OGRE. Experimental resultsshow that the virtual human has high fidelity and is easy for visual comparison.At last, to solve matching problem between two unequal-length motion time series,we adopt dynamic time warping algorithm to analyze the difference between the studentmotion and the standard motion and then can give right score when the student learn amotion.