Karate Video Reconstruction to Observe Player Motion and Flow of Competition

Abstract

It is well known that sport skill learning for trainees is facilitated via video observation of both the players’ actions and the overall flow of competition on the playing field. A video zoomed in on a player’s actions is suitable for learning the technical skills of the player, and a video showing all the players on the competition court is suitable for learning tactical skills such as player formation. The purpose of this study is to establish a method that can make narrow field-of-view videos available for observing the player positioning throughout the competition court. This paper focuses on karate competition videos and proposes an image processing method that extracts the image of the player area from the video frame and superimposes it on a standard karate court model image. Before the superimposition, the court model image is homographically transformed so that it fits the court in the video image. Using this approach, a player-focused video is reconstructed as a video of the entire competition field. For the reconstruction, a vertex matching method is developed for homography matrix calculation, and another calculation method for homography matrix calculation using only two straight lines is also developed for images that lack vertex information. Finally, an experimental result shows that a karate player-focused video is transformed successfully into a video of the entire karate competition field. Future work will focus on experimentally verifying the effectiveness of the method to the skill learning.