Arbitrary view image generation using a single camera based on rectangular parallelepiped approximation and known environment information

Summary form only given. Various methods for generating arbitrary view images from input images of cameras have been proposed. These conventional methods used multiple cameras. In this paper, we propose a method to generate arbitrary view images using a single camera in the known environment. We use one pan tilt camera. By using the camera parameters and 3 dimensional (3D) spatial information of the background, we estimate the 3D model of the unknown object in the environment. In particular, by recognizing the object based on the rectangular parallelepiped approximation, we achieve the arbitrary view image generation of diversiform objects and moving objects. In some simulations, we show the effectivity of the proposed method.