A Comprehensive Survey of Neural Radiance Fields for 3DScene Reconstruction

Abstract

Three-dimensional (3D) scene reconstruction is a long-standing problem incomputer vision, with applications in augmented/virtual reality, autonomous robotics, andcomputer graphics. Recently, Neural Radiance Fields (NeRFs) have emerged as a powerfulnew paradigm for 3D representation and novel view synthesis. This literature review pro-vides an overview of both classical 3D reconstruction methods and modern neural renderingapproaches, with equal emphasis on each. We cover the foundations of multi-view 3D recon-struction (structure-from-motion and multi-view stereo) alongside the formulation of NeRFsand implicit volumetric representations. We then survey advances in neural rendering andvolumetric scene representations, including efficient NeRF variants (such as Instant NeuralGraphics Primitives and PlenOctrees) and techniques for real-time rendering. Integrationof NeRFs with SLAM and robotics is discussed, highlighting how neural representationsare being combined with simultaneous localization and mapping. Benchmark datasets andevaluation metrics common to both domains are summarized. The review is organized in anIEEE conference style, with sections on Introduction, Methodologies (classical and neural),Discussion of current challenges, and Conclusion. A comprehensive reference list in IEEEformat is provided.