A Background Modeling Method with Simple Operations for 3D Video

This paper proposes a novel 3D background modeling method that only needs simple manual operations. For generating a background 3D model, the background area of a 3D scene is divided into a "ground region" and a "distant region." The 3D model of the distant region is represented by vertical rectangular planes with video texture. The ground region is modeled as a simple horizontal plane. These two regions are specified by a few points given manually in each image of multiple cameras. After the points are given by an operator, operation of the 3D background modeling is conducted based on the knowledge of camera poses and the target scene. We generated 3D videos of a background by using multiple 2D videos recorded at the Japan National Kasumigaoka Stadium and the Gymnasium of the Japan Institute of Sports Science to show the validity of our method.