In the field of 3D reconstruction, there has been a notable shift from traditional Structure-from-Motion methods to Neural Radiance Fields (NeRF). Introduced in 2020, NeRF has quickly gained global attention for its ability to reconstruct complex 3D scenes using only posed 2D images, achieving impressive synthesis results. Its effectiveness, particularly in rendering and synthesizing large-scale scenes, underscores its growing significance. This review examines NeRF’s applications and developments in large-scale scene reconstruction over recent years. We explore its specific applications, enhancements, development directions, and performance comparisons among key NeRF models. Our goal is to introduce NeRF to a broader research audience, providing a comprehensive reference for significant work in this field and inspiring future research directions.