Neural radiance fields (NeRFs) refer to a suit of deep neural networks that are used to learn and represent objects or scenes. Generally speaking, NeRFs have five main characters: volumetric rendering, novel view synthesis, factorizable embedded space, multi-view consistency and weighted importance sampling. Recently, NeRFs have drawn great attention and are now important cornerstones of metaverse and augmented reality research, as {is} their stronger efficiency and more imaginative rendering performance. There have been many reviews of NeRFs, most of them focus on different applications of NeRFs. In this paper, we provide a deep review and analysis of recent NeRF related works, according to the main characters of NeRFs they make further progress in. Then we introduce some new application innovations of NeRFs, and illustrate future opportunities of them. We hope this paper can provide an insightful organization of current developments in NeRFs, identify their limitations, and give suggestions for further research.