1 College of Electric Power, Inner Mongolia University of Technology, Hohhot 010080, China
2 Key Laboratory of Mechanical and Electrical Control at Universities of Inner Mongolia Autonomous Region, Hohhot 010080, China
3 Key Laboratory of Smart Control for New Energy Power System of Inner Mongolia Autonomous Region, Hohhot 010080, China
Show less
文章历史+
Received
Published
Revised
2025-08-20
2026-08-08
2026-01-09
Online Date
2026-09-03
PDF (24188K)
摘要
为解决传统视觉同步定位与建图方法在动态环境中易受运动目标干扰、鲁棒性弱、定位精度低的问题,提出了一种目标检测与几何约束融合的动态环境视觉同步定位与地图构建(simultaneous localization and mapping,SLAM)方法(YOLO-GK-SLAM)。首先,在ORB-SLAM3框架基础上,设计了一种语义目标检测与多视几何约束深度融合的动态特征点剔除机制,实现了动态特征点精确剔除与动态区域内静态信息的重识别。其次,为提升运行效率,提出了一种精细化关键帧筛选机制,压缩了候选关键帧数量,优先保留了静态结构的重点标志信息,提升了回环检测与位姿优化的运行效率。在此基础上,构建了一种融合关键帧图像与深度信息的动态场景稠密建图方法,提升了系统在动态环境下的环境建模能力。在TUM动态数据集上的实验表明,YOLO-GK-SLAM方法的绝对轨迹误差、相对位姿误差较经典方法分别至少降低了14.34%、17.48%,同时提升了地图的完整性及其边界的清晰度。在真实场景测试中,该方法依然保持了较高的定位精度,显著提升了系统在动态环境下的鲁棒性与定位稳定性。
Abstract
In order to mitigate moving-object interference and improve robustness and localization accuracy of visual SLAM (simultaneous localization and mapping) systems in dynamic environments, a visual SLAM method integrating object detection and geometric constraints, termed YOLO-GK-SLAM, was used to investigate accurate localization and reliable mapping in dynamic scenes. First, based on the ORB-SLAM3 framework, a dynamic feature point removal mechanism integrating semantic object detection with multi-view geometric constraints was designed to achieve accurate elimination of dynamic feature points and re-identification of static information within dynamic regions. Second, a refined keyframe selection strategy was proposed to reduce candidate keyframes while preserving salient static landmarks, thereby improving the efficiency of loop closure detection and pose optimization. The results show that, on the TUM dynamic dataset, the proposed YOLO-GK-SLAM method reduces ATE (absolute trajectory error) and RPE (relative pose error) by at least 14.34% and 17.48%, respectively, compared with classical visual SLAM methods, while significantly improving map completeness and boundary clarity. It is concluded that the proposed method maintains high localization accuracy and stability in real-world dynamic environments, effectively alleviates the influence of moving objects, and significantly enhances the robustness and practical applicability of visual SLAM systems.
GENG Yu-long, KOU Zhi-wei, LIU Bo, CUI Xiao-ming, QI Yong-sheng.
Visual SLAM Method for Dynamic Environments Based on Fusion of Object Detection and Geometric Constraints[J].
Science Technology and Engineering, 2026, 26(22): 9562-9575 DOI:10.12404/j.issn.1671-1815.2506077
在回环检测与全局优化部分,系统沿用了ORB-SLAM3的整体设计。通过数据库查询实现回环候选帧的快速匹配,并结合Sim3/SE3(similarity transformation in 3D/ special euclidean group in 3D)计算与闭环融合,进一步减少累计漂移误差。全局BA优化在整个系统的后端执行,用于保持全局轨迹与地图的一致性,并在回环闭合后完成地图更新。
ZhuJ, LingH, ChenK, et al. NICE-SLAM: neural implicit scalable encoding for SLAM[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024, 46(3): 1543-1558.
[2]
ZhangZ, LiuS, DengB, et al. Co-SLAM: joint coordinate and sparse-view optimization for neural implicit SLAM[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Vancouver: IEEE, 2023: 1332-1342.
[3]
CaoA Q, SunX, WangP, et al. Point-SLAM: dense neural point cloud SLAM[J]. IEEE Robotics and Automation Letters, 2023, 8(4): 2333-2340.
[4]
TeedZ, DengJ. DROID-SLAM: deep visual SLAM for dynamic environments[C]// Advances in Neural Information Processing Systems. Online: Curran Associates, Inc., 2021, 34: 26618-26629.
LiYingxin, ZuoTao, ZhaoXiong. Semi-dense semantic map construction of monocular vision based on object detection[J]. Science Technology and Engineering, 2023, 23(15): 6495-6505.
[7]
LuX, WangH, TangS, et al. DM-SLAM: monocular SLAM in dynamic environments[J]. Applied Sciences, 2020, 10(12): 4252.
YangYunxiu, HanBaoling, LuoXiao. Indoor navigation map research of simultaneous localization and mapping for mobile robots[J]. Science Technology and Engineering, 2021, 21(18): 7597-7603.
WangXuan, HouHewei, FuYinfang, et al. Optimization of mapping and localization system of quadruped robot for converter station inspection[J]. Science Technology and Engineering, 2023, 23(21): 9073-9081.
[12]
SaputraM R U, MarkhamA, TrigoniN. Visual SLAM and structure from motion in dynamic environments: a survey[J]. ACM Computing Surveys, 2018, 51(2): 1-36.
[13]
WangK, MaS, ChenJ, et al. Approaches, challenges, and applications for deep visual odometry: toward complicated and emerging areas[J]. IEEE Transactions on Cognitive and Developmental Systems, 2020, 14(1): 35-49.
[14]
Wan AasimW F A, OkashaM, FarisW F. Real-time artificial intelligence based visual simultaneous localization and mapping in dynamic environments: a review[J]. Journal of Intelligent and Robotic Systems, 2022, 105(1). DOI: 10.1007/s10846-022-01655-8.
WangFaxian, ShiZhonghua, ZhangKui. Track inspection recognition technology at turnout environment based on YOLOv8[J]. Science Technology and Engineering, 2025, 25(13): 5476-5483.
WangShuang, LiuYunping, ZhangBingqi, et al. Dynamic SLAM method based on panoptic segmentation and multi-view geometry[J]. Electronic Measurement Technology, 2024, 47(24): 149-159.
YangGang, ZhangWengang, CaoTianle. RGB-D SLAM with pose decoupling based on point-line-plane features[J]. Optics and Precision Engineering, 2025, 33(6): 993-1006.
WangHongxing, HeWenlong, WangJingyuan, et al. Real-time semantic visual SLAM for mobile robots in dynamic scenes[J]. Journal of Ordnance Equipment Engineering, 2025, 46(4): 252-262.
[23]
BescosB, FácilJ M, CiveraJ, et al. DynaSLAM: tracking, mapping, and inpainting in dynamic scenes[J]. IEEE Robotics and Automation Letters, 2018, 3(4): 4076-4083.
[24]
HeK M, GkioxariG, DolláRP, et al. Mask R-CNN[C]// IEEE International Conference on Computer Vision (ICCV). Venice: IEEE, 2017: 2980-2988.
[25]
YuC, LiuZ X, LiuX J, et al. DS-SLAM: a semantic visual SLAM towards dynamic environments[C]// IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Madrid: IEEE, 2018: 1168-1174.
[26]
BadrinarayananV, KendallA, CipollaR. SegNet: a deep convolutional encoder-decoder architecture for image segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(12): 2481-2495.
[27]
LiuY B, MiuraJ. RDS-SLAM: real-time dynamic SLAM using semantic segmentation methods[J]. IEEE Access, 2021, 9: 23772-23785.
[28]
WuW X, GuoL, GaoH L, et al. YOLO-SLAM: a semantic SLAM system towards dynamic environment with geometric constraint[J]. Neural Computing and Applications, 2022, 34(8): 6011-6026.
[29]
ZhangX G, ZhangR D, WangX K. Visual SLAM mapping based on YOLOv5 in dynamic scenes[J]. Applied Sciences, 2022, 12(22): 11548.
[30]
CamposC, ElenaR, GarcíaJ J, 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.
[31]
LeiM, WangX, LiJ, et al. YOLOv13: real-time object detection with hyper-graph-enhanced adaptive visual perception[J]. arXiv Preprint, arXiv: 2506.17733, 2025.
[32]
FengY, ZhangZ, WuQ, et al. Hyper-YOLO: when visual object detection meets hypergraph computation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025, 47(4): 2388-2401.
[33]
JaniM, FayyadJ, Al-YounesY, et al. Model compression methods for YOLOv5: a review[J]. arXiv Preprint, arXiv: 2307.11904, 2023.
ZhaoWei, WangFeng, MaXingyu, et al. Visual SLAM algorithm based on dynamic region removal and dense map construction[J]. Acta Armamentarii, 2025, 46(3): 304-314.