PulseExploreJournal ClubDebatesTrendingResearchersJournals
Instagram
HomeExploreJournal ClubTrending
Synapse
⌘+K
Synapse
May 20, 2026Journal of Computer-Aided Design & Computer Graphics0 citationsOpen Access

A Review of Dynamic Scene Reconstruction Based on 3D Gaussian Splatting

View Full Paper
ZLZhenhan LianZWZhangye WangJNJiong Niu

Key Points

  • The aim is to review 3D Gaussian splatting techniques for reconstructing dynamic scenes from images and their applications.
  • Categorized dynamic scene reconstruction methods into three types: deformation fields, spatio-temporal 4D Gaussian bodies, and inter-frame dynamics transmission.
  • Discussed the basic principles and advantages of 3D Gaussian splatting.
  • Analyzed the pros and cons of each method regarding scene representation and motion modeling.
  • Outlined current advancements in dynamic scene reconstruction.
  • Identified potential future research areas such as designing spatio-temporal data structures and large-scale dynamic scene rendering.

Abstract

作为计算机图形学与视觉领域的重点研究方向,动态场景重建的核心目标是从单目或多视角的二维图像序列或视频中精确地捕获并恢复动态场景的三维几何结构及其时变外观属性。三维高斯泼溅(3D Gaussian splatting,3DGS)是一种基于三维高斯体的场景重建和渲染技术,高斯体的参数化特性使其既可以显式地表示场景元素,又可以隐式地优化局部细节,从而高效地模拟场景中的光线传播和散射,最终实现实时渲染。文中对基于3DGS的动态场景重建方法进行梳理,并将其分为3大类。首先介绍3DGS的基本原理和优势,并简述相关的应用;然后总结基于变形场、时空四维高斯体和帧间动态传递3类方法,并围绕场景表示、运动建模、训练策略等维度展开描述,通过对比分析了每类方法的优缺点;最后回顾该领域的发展现状,指出未来可能的研究方向,包括时空数据结构的设计、稀疏视角下的动态重建,以及大规模动态场景的实时重建。

Ask AI
Helpful
Bookmark
Share
View Full Paper

Cite This Study

Lian et al. (2025) studied this question.

synapsesocial.com/papers/6a0d5013f03e14405aa9bac2https://doi.org/10.3724/sp.j.1089.2025-00251
Ask AI
Helpful
Bookmark
Share
View Full Paper