4D Dynamic Scene Reconstruction: A Comprehensive Survey

Ziren Gong, Guo Chen, Yongjia Li, Yihua Shao, Fabio Tosi, Stefano Mattoccia, Matteo Poggi, Hao Tang, Fei Ma, Shuyan Li, Ziyang Yan, Nicu Sebe, Ling Shao, Jianfei Cai, Qi Tian, Ming-Hsuan Yang

Abstract:

4D scene reconstruction in dynamic environments is a fundamental problem in robotics and computer vision, as it aims to
model scenes that evolve over time. Traditional geometric pipelines face inherent challenges in handling non-rigid motion, occlusions,
appearance variations, and maintaining temporal consistency. Recent advances in neural radiance fields, including Neural Radiance
Fields (NeRF) and 3D Gaussian Splatting (3DGS), have significantly pushed the boundaries by enabling high-fidelity, continuous, and
real-time 4D reconstruction beyond the capabilities of classical approaches. Despite this progress, a comprehensive survey dedicated to
radiance field-based 4D reconstruction remains lacking. This paper presents a systematic review of recent developments, with a focus on
NeRF- and 3DGS-based methods. We introduce a unified taxonomy and conceptual pipelines for existing 4D reconstruction systems,
and provide a critical analysis of their strengths and limitations. In addition, we summarize representative datasets and evaluation
metrics, and discuss key open challenges that warrant further investigation. This survey aims to offer a coherent and structured
foundation for advancing 4D reconstruction in dynamic and complex real-world environments. A continuously updated version is available
online at https://github.com/ZiyangYan/Awesome-4D-Scene-Reconstruction