北京信息科技大学学报（自然科学版）2025，Vol.40Issue(6)：49-58,10.DOI:10.16508/j.cnki.11-5866/n.2025.06.006
基于3D高斯泼溅与语义分割的室内场景三维重建
Indoor scene 3D reconstruction based on 3D Gaussian splatting and semantic segmentation
摘要
Abstract
3D scene reconstruction technology is one of the key techniques for applications such as robot navigation and virtual reality.Existing simultaneous localization and mapping(SLAM)algorithms based on 3D Gaussian splatting(3DGS)adopt unified rendering strategies for all objects and thus cannot accurately represent the visual effects of special optical materials such as glass and screens,resulting in limited visual quality of scene reconstruction.To address this problem,a SLAM algorithm based on 3DGS and semantic segmentation was proposed.Special objects like glass and screens were automatically identified through joint inference of vision foundation models,and a 3D object label mapping strategy was designed to convert 2D identification results into 3D space.Meanwhile,a joint optimization loss function incorporating multimodal constraints was designed to improve the rendering realism of special objects while maintaining the geometric reconstruction accuracy and localization performance of the SLAM algorithm.Experimental results demonstrate that the proposed algorithm outperforms existing methods in rendering quality for glass and screen regions while maintaining good localization accuracy.关键词
三维重建/同步定位与地图构建(simultaneous localization and mapping,SLAM)/语义分割/3D高斯泼溅Key words
3D reconstruction/simultaneous localization and mapping(SLAM)/semantic segmentation/3D Gaussian splatting(3DGS)分类
信息技术与安全科学引用本文复制引用
刘庭杉,刘伟,张洋,冯清娟,田咪..基于3D高斯泼溅与语义分割的室内场景三维重建[J].北京信息科技大学学报（自然科学版）,2025,40(6):49-58,10.基金项目
北京科艺空间文化有限公司横向项目(9162423320) （9162423320）
