College of Computer Science,Sichuan University,Chengdu 610065,China
Show less
文章历史+
Received
Accepted
Published
2025-11-19
2025-12-29
2026-05-28
Issue Date
2026-06-11
PDF (1725K)
摘要
传统的RGB-D视觉SLAM(Visual Simultaneous Localization and Mapping, vSLAM)系统主要依赖于单一特征或简单的组合特征进行定位和建图。然而,此类方法往往无法实现较高的定位准确度,并且在纹理少、噪声大的场景中构建出的地图精度较差。相比之下,多特征融合在降低定位误差方面展示出卓越的性能,同时还能增强系统的鲁棒性。因此,本文提出了一种新颖的基于点-线-面特征融合并带有假定平面约束的RGB-D视觉SLAM系统(SVPLP-SLAM)。该系统集成点特征、线特征和平面特征进行定位和建图,通过使用RGB-D相机从彩色图像中提取点特征和线特征,从深度图像中提取平面特征,采用关联匹配和长度抑制策略,剔除无效的长距离短线特征,使用假定平面作为边缘约束,获得具有正交结构的平面特征。线特征和平面特征的引入可以有效减少跟踪模块所需的特征数量,从而降低光束法平差 (Bundle Adjustment, BA)优化的计算复杂度。最后,采用绝对轨迹误差(Absolute Trajectory Errors, ATEs)的均方根误差(Root Mean Squate Error, RMSE)作为依据,在TUM RGB-D和ICL-NUIM数据集上的实验结果证明,在定位精度上,该方法优于当前主流方法。
Abstract
Traditional RGB-D Visual Simultaneous Localization and Mapping (vSLAM) systems rely predominantly on a single kind of feature or a simplistic combination of features.However, such approaches often fail to achieve optimal accuracy and mapping completeness, particularly in scenarios characterized by low texture and significant noise.In contrast,the fusion of multiple features has shown superior performance in mitigating localization inaccuracies while simultaneously enhancing robustness.Therefore, this paper proposes a novel RGB-D Visual SLAM system Using Point-Line-Plane Feature Fusion with Supposed Plane Constraints (SVPLP-SLAM) that synergistically integrates point features, line features, and plane features for localization and mapping.The system uses an RGB-D camera to extract point features and line features from the RGB images, and plane features from the depth images.By employing correlation matching and length suppression strategies, invalid long-distance short line features can be eliminated.Using supposed planes as edge constraints, the system can obtain plane features with orthogonal structures.The introduction of line features and plane features can effectively decrease the number of features required by the tracking module, thereby reducing the computational complexity of Bundle Adjustment (BA) optimization.Finally, using the root mean square error (RMSE) of absolute trajectory errors (ATEs) as a basis, extensive experiments on the TUM RGB-D and ICL-NUIM datasets verify that the proposed method outperforms current mainstream methods in terms of localization accuracy.
其中,与是两个标量参数,用于对直线方向进行低维参数化。 U 是三维正交旋转矩阵,用于将局部参数化结果映射到世界坐标系下的三维空间。 W 是具有旋转-缩放结构的二维矩阵,其缩放因子为:,用于在优化过程中保持方向表示的结构一致性,并避免方向发生退化。设R( θ )= U,R(α)= W,则对应相关的旋转变换计算方式为:
TaketomiT, UchiyamaH, IkedaS.Visual SLAM algorithms: A survey from 2010 to 2016 [J].IPSJ Trans Comput Vis Appl, 2017, 9(1): 1-11.
[2]
WangP, ZhaoL, ZhangY, et al.MBA-SLAM: Motion blur aware dense visual SLAM with radiance fields representation [J].IEEE Trans Pattern Anal Mach Intell, 2025, 47(12): 11168-11186.
[3]
WhelanT, LeuteneggerS, Salas MorenoR, et al.ElasticFusion: Dense SLAM without a pose graph [C]//Robotics: Science and Systems Foundation, 2015: 3.
[4]
JohariM M, CartaC, FleuretF.ESLAM: Effi-cient dense SLAM system based on hybrid representation of signed distance fields [C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023: 17408-17419.
[5]
LeeJ, ParkS Y.PLF-VINS: Real-time monoc-ular visual-inertial SLAM with point-line fusion and parallel-line fusion [J].IEEE Robot Autom Lett, 2021, 6(4): 7033-7040.
[6]
XuL, YinH, ShiT, et al.EPLF-VINS:Real-time monocular visual-inertial SLAM with efficient point-line flow features [J].IEEE Robot Autom Lett, 2023, 8(2): 752-759.
[7]
von GioiR G, JakubowiczJ, MorelJ M, et al.LSD: A fast line segment detector with a false detection control [J].IEEE Trans Pattern Anal Mach Intell, 2010, 32(4): 722-732.
[8]
WangX, ChristieM, MarchandE.TT-SLAM: Dense monocular SLAM for planar environ-ments [C]//2021 IEEE International Conference on Robotics and Automation (ICRA), 2021: 11690–11696.
[9]
ChenD, WangS, XieW, et al.VIP-SLAM: An efficient tightly-coupled RGB-D visual inertial planar SLAM [C]//2022 International Confer-ence on Robotics and Automation (ICRA), 2022: 5615-5621.
[10]
ShuF, XieY, RambachJ, et al.Visual SLAM with graph-cut optimized multi-plane recon-struction [C]//2021 IEEE International Sympo-sium on Mixed and Augmented Reality Adjunct (ISMAR-Adjunct), 2021: 165-170.
[11]
HoiemD, EfrosA A, HebertM.Geometric context from a single image [C]//Tenth IEEE International Conference on Computer Vision (ICCV'05), 2005: 654-661.
[12]
Mur-ArtalR, TardosJ D.ORB-SLAM2: An open-source SLAM system for monocular, ste-reo, and RGB-D cameras [J].IEEE Trans Robot, 2017, 33(5): 1255-1262.
[13]
CamposC, ElviraR, RodriguezJ J G, et al.ORB-SLAM3: An accurate open-source library for visual, visual-inertial, and multimap SLAM [J].IEEE Trans Robot, 2021, 37(6): 1874-1890.
[14]
SuX, EgerS, MisikA, et al.HPF-SLAM: An efficient visual SLAM system leveraging hybrid point features [C]//2024 IEEE International Conference on Robotics and Automation (ICRA), 2024: 15929-15935.
[15]
BirdJ, BlumenkampJ, ProrokA.DVM-SLAM: Decentralized visual monocular simultaneous localization and mapping for multi-agent sys-tems [C]//2025 IEEE International Conference on Robotics and Automation (ICRA), 2025: 1-7.
[16]
LemaireT, LacroixS.Monocular-vision based SLAM using line segments [C]//2007 IEEE In-ternational Conference on Robotics and Auto-mation (ICRA), 2007: 2791-2796.
[17]
Gomez-OjedaR, MorenoF A, Zuniga-NoelD, et al.PL-SLAM: A stereo SLAM system through the combination of points and line segments [J].IEEE Trans Robot, 2019, 35(3): 734-746.
[18]
KaessM.Simultaneous localization and map-ping with infinite planes [C]//2015 IEEE Inter-national Conference on Robotics and Automa-tion (ICRA), 2015: 4605-4611.
[19]
LuskP C, ParikhD, HowJ P.GraffMatch: Global matching of 3D lines and planes for wide baseline LiDAR registration [J].IEEE Robot Autom Lett, 2023, 8(2): 632-639.
[20]
YangH, YuanJ, GaoY, et al.UPLP-SLAM: Unified point-line-plane feature fusion for RGB-D visual SLAM [J].Inf Fusion, 2023, 96: 51-65.
[21]
WangX, ZhangY, ZhangZ, et al.FI-SLAM: Feature fusion and instance reconstruction for neural implicit SLAM [C]//2024 IEEE/RSJ In-ternational Conference on Intelligent Robots and Systems (IROS), 2024: 527-532.
[22]
PapatheodorouS, BocheS, LainaS B, et al.Efficient submap-based autonomous MAV ex-ploration using visual-inertial SLAM configu-rable for LiDARs or depth cameras [C]//2025 IEEE International Conference on Robotics and Automation (ICRA), 2025: 7284-7290.
[23]
ZhangX, WangW, QiX, et al.Point-plane SLAM using supposed planes for indoor envi-ronments [J].Sensors, 2019, 19(17): 3795.
[24]
ZhangG, LeeJ H, LimJ, et al.Building a 3-D line-based map using stereo SLAM [J].IEEE Trans Robot, 2015, 31(6): 1364-1377.
[25]
HartleyR, ZissermanA.Multiple view geom-etry in computer vision [M].2nd ed.Cambridge, USA: Cambridge University Press, 2003.
[26]
FuQ, WangJ, YuH, et al.Pl-vins: Real-time monocular visual-inertial slam with point and line [PP/OL].V3.arxiv(2022-04-15)[2025-09-25].
[27]
LeeS J, HwangS S.Elaborate monocular point and line SLAM with robust initializa-tion [C]//Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision (ICCV), 2019.
[28]
ShuF, WangJ, PaganiA, et al.Structure PLP-SLAM: Efficient sparse mapping and lo-calization using point, line and plane for mo-nocular, RGB-D and stereo cameras [C]//2023 IEEE International Conference on Robotics and Automation (ICRA).London, United Kingdom: IEEE, 2023: 2105-2112.
[29]
ZhangL, KochR.An efficient and robust line segment matching approach based on LBD de-scriptor and pairwise geometric consistency [J].J Vis Commun Image Represent, 2013, 24(7): 794-805.
[30]
TrevorA, GedikliS, RusuR B, et al.Efficient organized point cloud segmentation with con-nected components [C]//2013 Semantic Perception Mapping and Exploration (SPME), 2013: 1-6.
[31]
MaL, KerlC, StucklerJ, et al.CPA-SLAM: Consistent plane-model alignment for direct RGB-D SLAM [C]//2016 IEEE International Conference on Robotics and Automation (ICRA), 2016: 1285-1291.
[32]
Mur-ArtalR, TardosJ D.Fast relocalisation and loop closing in keyframe-based SLAM [C]//2014 IEEE International Conference on Robotics and Automation (ICRA), 2014: 846-853.
[33]
Galvez-LópezD, TardosJ D.Bags of binary words for fast place recognition in image se-quences [J].IEEE Trans Robot, 2012, 28(5): 1188-1197.
[34]
KummerleR, GrisettiG, StrasdatH, et al.G2o: A general framework for graph optimiza-tion [C]//2011 IEEE International Conference on Robotics and Automation, 2011: 3607-3613.
[35]
SturmJ, EngelhardN, EndresF, et al.A benchmark for the evaluation of RGB-D SLAM systems [C]//2012 IEEE/RSJ International Con-ference on Intelligent Robots and Systems, 2012: 573-580.
[36]
HandaA, WhelanT, McDonaldJ, et al.A benchmark for RGB-D visual odometry, 3D reconstruction and SLAM [C]//2014 IEEE In-ternational Conference on Robotics and Auto-mation (ICRA), 2014: 1524-1531.
[37]
ZhangC.PL-GM: RGB-D SLAM with a novel 2D and 3D geometric constraint model of point and line features [J].IEEE Access, 2021, 9: 9958-9971.
[38]
YunusR, LiY, TombariF.ManhattanSLAM: Robust planar tracking and mapping leveraging mixture of Manhattan frames [C]//2021 IEEE International Conference on Robotics and Au-tomation (ICRA), 2021: 6687-6693.
[39]
BamdadM, HutterH-P, DarvishyA.Deep learning-powered visual SLAM aimed at as-sisting visually impaired naviga-tion [C]//Proceedings of the 20th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applica-tions, 2025: 758-765.
[40]
YangG, LiuP, MaW.Enhanced RGB-D SLAM through orthogonal plane constraints and point-line-plane collaborative optimization [J].PLoS One, 2025, 20(9): e0330839.
[41]
EndresF, HessJ, SturmJ, et al.3-D mapping with an RGB-D camera [J].IEEE Trans Robot, 2014, 30(1): 177-187.
[42]
SunQ, YuanJ, ZhangX, et al.RGB-D SLAM in indoor environments with STING-based plane feature extraction [J].IEEE/ASME Trans Mechatron, 2018, 23(3): 1071-1082.
[43]
XieY, ShuF, RambachJ,et al.PlaneRecNet: Multi-task learning with cross-task consistency for piece-wise plane detection and reconstruc-tion from a single rgb image [PP/OL].V2.arxiv(2022-01-30)[2025-09-25].
[44]
XiaoX, ChenQ S, WuS Y,et al.Research on the application of deep learning methods in identification localization and recognition of engineering drawings[J].Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition),2025,37(5):696-707.