Conventional captured-image-based virtual reality (VR) systems only support rotational view direction changes, roll, yaw, and pitch. It is 3-degree-of-freedom (3-DoF). DoF represents the user’s movements, and the highest DoF is 6, which includes thr...
Conventional captured-image-based virtual reality (VR) systems only support rotational view direction changes, roll, yaw, and pitch. It is 3-degree-of-freedom (3-DoF). DoF represents the user’s movements, and the highest DoF is 6, which includes three rotational view direction changes, roll, yaw, and pitch, as well as three translational viewpoint movements along the x, y, and z axes. The limited DoF of the conventional captured-image-based VR lowers user’s sense of reality.
Light field (LF), which can generate a view at a free viewpoint through a combination of light rays, is a suitable approach to support the freedom to change viewpoints. LF assumes a planar or spherical surface, and generates a view by combing light rays passing through the surface. In particular, the spherical LF system creates a 360-degree view through light rays incident from 360-degree direction. In the spherical LF, a viewpoint freely moves along the x, y, and z axes inside the sphere and changes view direction, and thus 6-DoF is supported. However, it is difficult to acquire light rays for LF assuming a planar or spherical surface. In the case of spherical LF, a special equipment for rotating multiple cameras arranged in an arch shape is used to acquire light rays along a spherical surface. In order to cover a larger space, it is necessary to assume a larger plane or spherical surface. The larger the surface, the more difficult the light ray acquisition is.
3D LF consists of light rays acquired along the line, unlike conventional LF that assumes surfaces. A line instead of a plane and a circular structure instead of a spherical surface are used to construct 3D LF. It is easy to acquire light rays, which is acquired by moving the camera mounted on a camera slider and a dolly along the line. However, 3D LF cannot acquire vertical parallax because it obtains light rays at only one vertical point, and it causes distortion of the generated views. Assuming a larger structure does not significantly increase the difficulty of acquisition, but it increases the distortion of 3D LF view generation.
This paper aims to develop a free viewpoint VR system for large space based on 3D LF. In contrast to extending the structure in the existing method, it assumes a 3D LF Stack in which multiple 3D LFs are stacked in front and back. The proposed system is simple to obtain light rays and limits the distortion to a certain range. In addition, two 3D LF Stacks are arranged orthogonally to generate a 360-degree view at a free viewpoint. There are two challenges for the proposed system. First is the need to connect independent 3D LFs. The existing LF-based approach creates a view using a single LF, while the proposed system generates a view using four 3D LFs. This paper proposes two 3D LF connection methods and introduces appropriate usage methods according to various implementation environments.
Another is that 3D LF Stack still contains distortion, and the error is particularly noticeable as the viewpoint moves and the 3D LF that generates a view changes. This paper proposes a view generation method using a light ray set with epipolar geometry relationship in 3D LF stack.