Recently, Neural Radiance Fields (NeRF) have been attracting attention as a method for arbitrary viewpoint image generation, because NeRF enables ray rendering by modeling a 3D space with neural fields. On the other hand, however, learning and rendering a large scene with NeRF is difficult and computationally expensive. In particular, scenes related to omnidirectional images, such as those used in this study, are outdoor or whole indoor scenes, which are wider than the scenes for a single object used in general NeRF, making these problems even more pronounced. Experimental results show that the proposed method has a wider range of viewpoint movement than the conventional method and is superior to NeRF in representing long distances. The computational cost of the proposed method is also lower than that of NeRF.