A 3-D Motion Recovery Technique for Group Sports Employing Uncalibrated Video Cameras

摘要

Human motion recovery and modeling is a demanding technique in various areas such as sports, dancing, virtual reality, video game producing, etc. But the people in this area focus their efforts more on motion recovery of a single person than a group of persons. In this paper, a technique is described for recovering 3-D motions of a group of persons who are engaged in a certain activity. More than three video cameras are placed apart in front of the persons concerned and their motions are captured. Commonly visible feature points which spread over them are tracked on the video image streams to yield an extended measurement matrix. This matrix receives singular value decomposition and a shape matrix is finally obtained which contains the 3-D locations of all the chosen feature points. The technique is applied to 3-D motion recovery of the players in a mini-basketball game performed in a gymnasium and satisfactory results are obtained. The present technique may have application to, e.g., formation evaluation of the players in a group sport as well as form analysis of individual players.