It is well known that sport skill learning is facilitated by video observation of players' actions in the target sport. A viewpoint change function is desirable when a learner observes the actions using video images. However, in general, viewpoint changes for observation are not possible because most videos are filmed from a fixed point using a single video camera. The objective of this research is to develop a method that generates a 3D human model of a player (i.e., a virtual player) from a single image and enable observation of the virtual player's action from any point of view. As a first step, this study focused on karate training and developed a semiautomatic method for 3D reconstruction from video images of sparring in karate.
展开▼