Automatic player's view generation of real soccer scenes based on trajectory tracking

Abstract

This paper proposes a method to generate a player's view video of an actual soccer match using a 3D free-viewpoint video technique. Users can enjoy 3D video simply by choosing a target player and concentrating on a soccer match from the player's view by manually controlling the viewing position. The generated video provides an immersive sight as if the user is running on the pitch. To generate the player views, the target player's 3D trajectory must be estimated. We developed a novel computer vision technique for player tracking that robustly works in an actual soccer stadium. In the current system, the orientation of the virtual camera for the player's view does not follow the gaze direction of each player because the image resolution is too poor to acquire gaze direction by computer vision. Users can choose a favorite orientation control method. We applied the proposed method to an actual soccer match held in an outdoor stadium to confirm its effectiveness.