GS-SLAM: a Real-time SLAM System Based on Semantic Enhancement and 3D Gaussian Splatting

LIU Yuqing; WANG Feng; HAN Xiaoming; SONG Lirui; LI Zeliang

doi:10.12382/bgxb.2025.1103

您当前的位置：

首页 >

文章列表页 >

GS-SLAM: a Real-time SLAM System Based on Semantic Enhancement and 3D Gaussian Splatting

更新时间：2026-07-31

- GS-SLAM: a Real-time SLAM System Based on Semantic Enhancement and 3D Gaussian Splatting
- Acta Armamentarii (2026)
- 作者机构：
  
  太原理工大学电气与动力工程学院,山西,太原,030024
- 作者简介：
- 基金信息：
- DOI：10.12382/bgxb.2025.1103
  CLC： TP391.41
- Received：16 December 2025，
  
  Online First：31 July 2026，
- 稿件说明：
移动端阅览
刘雨晴，王峰，韩晓明，等. GS-SLAM：基于语义增强与3D高斯溅射的实时SLAM系统[J/OL]. 兵工学报, 2026(2026-07-31). https://doi.org/10.12382/bgxb.2025.1103.

LIU Y Q, WANG F, HAN X M, et al. Gs-slam: a real-time slam system based on semantic enhancement and 3d gaussian splatting[J/OL]. Acta Armamentarii, 2026(2026-07-31). https://doi.org/10.12382/bgxb.2025.1103. (in Chinese)
刘雨晴，王峰，韩晓明，等. GS-SLAM：基于语义增强与3D高斯溅射的实时SLAM系统[J/OL]. 兵工学报, 2026(2026-07-31). https://doi.org/10.12382/bgxb.2025.1103. DOI：

LIU Y Q, WANG F, HAN X M, et al. Gs-slam: a real-time slam system based on semantic enhancement and 3d gaussian splatting[J/OL]. Acta Armamentarii, 2026(2026-07-31). https://doi.org/10.12382/bgxb.2025.1103. (in Chinese) DOI：

摘要

为了提升动态战场环境下无人平台的定位鲁棒性、场景重建质量与实时渲染能力。针对传统实时定位与建图（Simultaneous Localization and Mapping，SLAM）算法在动态环境中易出现定位漂移，以及稀疏点云或体素地图难以支持高质量3D可视化与渲染的双重挑战，提出一种融合鲁棒动态语义前端与3D高斯溅射建图后端的实时语义SLAM系统，GS-SLAM。前端采用基于改进MobileNetV2骨干网络的轻量化SOLOv2算法，实现实时语义分割；在此基础上，提出三帧极线一致性约束与退化场景下的双阈值几何约束，构建基于恒速运动假设与3D轮廓重投影的漏检补偿机制，实现对动态特征点的有效剔除。后端设计独立异步3D高斯建图线程，提出滑动窗口一致性动态掩码融合策略，对多帧语义和几何检测结果进行时序统计；构建融合匹配点数量与跟踪稳定性的自适应优选机制，筛选高质量关键帧，在动态场景中构建静态3D高斯模型，实现实时处理与高质量渲染。实验结果表明，在TUM数据集的高动态序列上，相比原始ORB-SLAM2，该算法的绝对轨迹误差显著降低（降幅最高超过99%），相对位姿误差降幅均达93%以上；在Bonn数据集上亦取得较优结果；相比同类先进算法具有更优的定位精度。

Abstract

This paper aims to enhance the localization robustness

scene reconstruction quality and real-time rendering capability of unmanned platforms in dynamic battlefield environments.To address the dual challengesoftraditionalsimultaneouslocalization andmapping(SLAM) algorithmsbeingprone to localization drift in dynamic environments anddifficultysupportinghigh-quality 3D visualization and renderingwithsparse point clouds or voxel maps

a real-time semantic SLAM system is proposed. This system integrates a robust dynamic semantic front-end with a 3D Gaussiansplatting mapping back-end.In the front-end

a lightweight SOLOv2 algorithm based on an improved MobileNetV2 backbonenetwork is used toachieve real-time semantic segmentation. To effectively eliminatethedynamic feature points

this paperproposesa three-frame epipolar consistency constraint and a dual-threshold geometric constraint for degraded scenes. These are supplemented by a missed detection compensation mechanism based on a constant velocity motion assumption and 3D contour reprojection.An independent asynchronous 3D Gaussian mapping threadis designed in theback-end.Asliding window consistency dynamic mask fusion strategyis proposed to conduct thetemporal statistical analysis onthemulti-frame semantic and geometric detection results. Furthermore

an adaptive selection mechanism based on the number of matched points and tracking stability is introduced to filter high-quality keyframes

thereby constructing a static 3D Gaussian model within dynamic scenes. Experimental results on the TUM dataset demonstrate that

compared to the original ORB-SLAM2

the proposed algorithm significantly reduces theabsolutetrajectoryerror (ATE) with a maximum reductionof more than99%

and consistently reduces therelativeposeerror (RPE) by over 93% in high-dynamic scenarios. Significant improvementsin trajectory estimation accuracyare also observed on the Bonn dataset.The proposed algorithm exhibitshassuperior localization accuracycompared tosimilarstate-of-the-art algorithms. Additionally

the systemrealizes real-time processing and high-quality rendering.

关键词

Keywords

references

蔡东伟, 邓中亮, 彭颖轩. 一种几何语义协同的动态视觉定位算法[C]//第十三届中国卫星导航年会.2022: 1-7.

CAI D W, DENG Z L, PENG Y X. A geometric-semantic collaborative dynamic visual localization algorithm[C]// The 13th China Satellite Navigation Conference. 2022: 1-7. (in Chinese).

MA Y J, LÜ J H, WEI J. High-precision visual SLAM for dynamic scenes using semantic-geometric feature filtering and NeRF maps[J].Electronics, 2025, 14(18): 3657.

DAVISON A J, REID I D, MOLTON N D, et al. MonoSLAM: real-time single camera SLAM[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2007, 29(6): 1052-1067.

CHUNG C M, TSENG Y C, HSU Y C, et al. Orbeez-SLAM: a real-time monocular visual SLAM with ORB features and NeRF-realized mapping[C]//2023 IEEE International Conference on Robotics and Automation. 2023: 9400-9406.

FORSTER C, ZHANG Z C, GASSNER M, et al. SVO: semidirect visual odometry for monocular and multicamera systems[J]. IEEE Transactions on Robotics, 2016, 33(2): 249-265.

CAMPOS C, ELVIRA R, RODRÍGUEZ J J G, et al.ORB-SLAM3: an accurate open-source library for visual, visual-inertial, and multimap SLAM[J].IEEE Transactions on Robotics, 2021, 37(6): 1874-1890.

LIU Y B, MIURA J. RDS-SLAM: real-time dynamic SLAM using semantic segmentation methods[J]. IEEE Access, 2021, 9: 23772-23785.

BARROS A M , MICHEL M, MOLINE Y, et al. A comprehensive survey of visual SLAM algorithms[J]. Robotics, 2022, 11(1): 24.

BESCOS B, CAMPOS C, TARDÓS J D, et al. DynaSLAM II: tightly-coupled multi-object tracking and SLAM[J]. IEEE Robotics and Automation Letters, 2021, 6(3): 5191-5198.

WANG X D, WANG X Y, GAO C, et al. Positioning accuracy optimization of DSO-SLAM and its use for the localization of unmanned vehicle[C]//2021 International Conference on Computer Communication and Artificial Intelligence. 2021: 155-160.

CHANG Z Y, WU H L, SUN Y L, et al. RGB-D visual SLAM based on YOLOv4-Tiny in indoor dynamic environment[J]. Micromachines, 2022, 13(2): 230.

LIU Q, YUAN J, KUANG B F. SIA-SLAM: a robust visual SLAM associated with semantic information in dynamic environments[J]. Multimedia Tools and Applications, 2024, 83(18): 53531-53547.

WANG X L, ZHANG R F, KONG T, et al. SOLOv2: dynamic and fast instance segmentation:arXiv:2003.10152[R].Ithaca,NY,US:Cornell University, 2020: 2003.10152.

LI J J, LUO J W. YS-SLAM: YOLACT++ based semantic visual SLAM for autonomous adaptation to dynamic environments of mobile robots[J]. Complex & Intelligent Systems, 2024, 10(4): 5771-5792.

BOLYA D, ZHOU C, XIAO F Y, et al. YOLACT: real-time instance segmentation:arXiv:1904.02689 [R].Ithaca,NY,US:Cornell University, 2019:1904.02689.

YE S P, XU B L, YANG Y, et al. DC-SLAM: dual-category dynamic feature suppression for RGB-D VSLAM[J]. Expert Systems with Applications, 2026, 299(Part A): 129952.

赵薇, 王峰, 马星宇, 等.基于动态区域剔除与稠密地图构建的视觉SLAM算法[J].兵工学报, 2025, 46(3): 240217.

ZHAO W, WANG F, MA X Y, et al. Visual SLAM algorithm based on dynamic region exclusion and dense map construction[J]. Acta Armamentarii, 2025, 46(3): 240217. (in Chinese).

REDMON J, FARHADI A. YOLO9000: better, faster, stronger[C]//Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Hawaii Convention Center, Hawaii, US: IEEE, 2017: 7263-7271.

翟伟光, 王峰, 马星宇, 等. YSG-SLAM:动态场景下基于YOLACT的实时语义RGB-D SLAM系统[J].兵工学报, 2025, 46 (6): 240443.

ZHAI W G, WANG F, MA X Y, et al. YSG-SLAM: a real-time semantic RGB-D SLAM based on YOLACT in dynamic scenes[J]. Acta Armamentarii, 2025, 46(6): 240443. (in Chinese)

YANG J M, WANG Y T, TAN X, et al. DHP-SLAM: a real-time visual SLAM system with high positioning accuracy under dynamic environment[J]. Displays, 2025, 89: 103067.

SUN Y, WANG Q, YAN C, et al. D-VINS: dynamic adaptive visual–inertial SLAM with IMU prior and semantic constraints in dynamic scenes[J]. Remote Sensing, 2023, 15(15): 3881.

LIU X T, ZHANG Y, LU G J, et al. DGO-VINS: a visual-inertial SLAM for dynamic environments with geometric constraint and adaptive state optimization[J]. IEEE Robotics and Automation Letters, 2025, 10(8): 8091-8098.

LI Y M, WANG Y Z, LU L W, et al. Semantic visual SLAM algorithm based on improved DeepLabV3+ model and LK optical flow[J]. Applied Sciences, 2024, 14(13): 5792.

VESPA E, FUNK N, KELLY P H J, et al. Adaptive-resolution octree-based volumetric SLAM[C]//2019 International Conference on 3D Vision. 2019: 654-662.

MATSUKI H, MURAI R, KELLY P H J, et al. Gaussian splatting SLAM[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR). 2024: 18039-18048.

ZHENG J H, ZHU Z, BIERI V, et al. WildGS-SLAM: monocular Gaussian splatting SLAM in dynamic environments:arXiv:2504.03886[R].Ithaca,NY,US:Cornell University, 2025: 2504.03886.

Views

下载量

CNKI被引量

Alert me when the article has been cited

提交

Tools

Publicity Resources

YSG-SLAM:a Real-time Semantic RGB-D SLAM Based on YOLACT in Dynamic Scene

Obstacle Recognition and Traversability Analysis of Tracked Vehicles in Off-road Environment

Autonomous UAV Location Selection Technique Based on Data Fusion

Visual SLAM Algorithm Based on Dynamic Region Exclusion and Dense Map Construction

A SLAM in Dynamic Environment Based on Instance Segmentation and Optical Flow

Related Author

Pengshuai MENG

Wei ZHAO

Xingyu MA

Feng WANG

Weiguang ZHAI

Yujian LIU

Yunlu TAN

Liang YU

Related Institution

College of Electronic and Information Engineering, Taiyuan University of Technology

School of Mechanical Engineering, Beijing Institute of Technology

Intelligent Systems Headquarters, China North Artificial Intelligence & Innovation Research Institute

College of Aerospace Engineering, Nanjing University of Aeronautics and Astronautics

Key Laboratory of Fundamental Science for National Defense-Advanced Design Technology of Flight Vehicle, Nanjing University of Aeronautics and Astronautics

Postal code：100089
Tel：010-68963060/68962718 Fax：010-68963025 Email：bgxb@cos.org.cn
Technical support is provided by Beijing Founder electronics co., LTD 京ICP备05059581号-4 京公网安备11010802024360号
It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
Total Visits：236583 Visits Today：898

⁰