School of Mechanical Engineering, Hebei University of Technology, Tianjin 300401, China
Show less
文章历史+
Received
Published
Revised
2025-04-29
2026-05-08
2025-11-28
Online Date
2026-08-06
PDF (16710K)
摘要
针对纯激光雷达即时定位与建图方法在大尺度复杂场景下存在精度衰减、鲁棒性欠佳的问题,提出一种基于因子图优化的紧耦合同时定位和建图技术(simultaneous localization and mapping,SLAM)方法。该方法通过三大核心设计提升性能:一是设计基于误差状态卡尔曼滤波并融合多传感器信息的组合惯导里程计因子;二是采用多尺度深度感知特征选择算法提取线面点特征,并通过先粗配准后精配准的方法优化帧间匹配,提升位姿变换精度;三是在因子图模型中整合组合惯导因子、激光惯性里程计因子,并引入具有旋转不变性的扫描上下文的回环检测因子,抑制累积误差。在KITTI公开数据集与校园自采数据集上的对比实验表明,所提算法相较于经典LIO-SAM算法,复杂场景下的绝对轨迹误差均值分别降低25.96%和23.01%,均方根误差分别降低25.52%和28.38%。实验验证了所提算法的高建图精度与鲁棒性,为室外复杂场景SLAM的工程应用提供了有效技术方案。
Abstract
In order to address the current limitations of pure laser SLAM under large-scale complex environments, particularly their lack of precision and poor robustness, a tightly-coupled SLAM method based on factor graph optimization has been proposed. The method enhances performance through three core designs. Firstly, designing an integrated inertial odometry factor based on the error-state Kalman filter (ESKF) that fuses multi-sensor information. Secondly, adopting a multi-scale depth-aware feature selection algorithm to extract line, plane, and point features, and optimizing inter-frame matching via coarse-to-fine registration to improve pose transformation accuracy. Thirdly, incorporating integrated inertial factors and laser-inertial odometry factors into the factor graph model, and introducing a rotation-invariant Scan Context-based loop closure detection factor to suppress cumulative errors. Comparative experiments on the public KITTI dataset and a self-collected campus dataset show that, compared with the classical LIO-SAM algorithm, the proposed algorithm reduces the mean absolute trajectory error (ATE) by 25.96% and 23.01% respectively, and the root mean square error (RMSE) by 25.52% and 28.38% respectively in complex scenarios. Experiments verify that the proposed algorithm achieves high mapping accuracy and robustness, and provides an effective technical solution for the engineering application of SLAM in outdoor complex scenarios.
DING Cheng-jun, MA Zhi-hui, ZHANG Tan, MA Teng-fei, LI Zi-jian, WANG Yi-tong, JING Zhi-kai.
Multi-scale Depth-aware Simultaneous Localization and Mapping Algorithm Based on Factor Graph Optimization[J].
Science Technology and Engineering, 2026, 26(13): 5556-5565 DOI:10.12404/j.issn.1671-1815.2503143
同时定位和建图技术(simultaneous localization and mapping,SLAM)已成为移动机器人及自动驾驶领域的一种重要技术手段[1],目前该技术主要分为基于视觉信息的方法和基于激光雷达信息的方法。作为感知环境和路径跟踪等功能的前置条件,需要在陌生环境下实时建立当前环境地图及获取自身位置信息,相比相机而言,雷达获得的信息更难受到环境光照和恶劣天气的影响,具有较强的鲁棒性。
针对LIO-SAM在全局轨迹绘制和回环检测方面的不足,现采用GTSAM(georgia tech smoothing and mapping)库实现因子图优化问题的建模。该算法通过对激光雷达、IMU、RTK等传感器数据的处理进而对车辆进行状态估计并建立地图信息。所提算法相对LIO-SAM算法主要的改进部分有以下内容:提出基于ESKF的组合惯导里程计因子,将高频IMU数据和GNSS全局定位信息通过紧耦合的方式优化估计的车辆位姿,从而实时修正位姿信息,降低定位误差;改进点云特征提取部分,通过多尺度深度感知的特征选择算法能够在大尺度场景中更准确地提取环境特征,提升建图准确性;通过改进点云帧间匹配部分获得更精确的空间位姿变换关系,避免在初始位姿不准确时算法陷入局部最优解的情况。通过在公开数据集和自采数据集上进行实验,验证所提算法。
为了处理系统的噪声及干扰,减少定位的不确定性,一般选择引入卡尔曼滤波器通过估计线性动态系统的状态来实现对定位估计值的优化。但由于在实际应用中,很多系统的观测模型都是非线性的,直接使用卡尔曼滤波器来估计系统的状态会产生较大的误差,因此提出引入误差状态卡尔曼滤波器(error state Kalman filter,ESKF)优化系统状态的预测值和误差状态预测。
YuLianxiang, ZhengMingkui, OuWenjun, et al. Optimization and system implementation of outdoor LiDAR SLAM algorithm for mobile robots with multi-sensor fusion[J]. Journal of Electronic Measurement and Instrumentation, 2023, 37(2): 48-55.
[3]
BeslP J, MckayH D. A method for registration of 3-D shapes[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 1992, 14(2): 239-256.
[4]
SegalA, HhnelD, ThrunS. Generalized-ICP[J]. Robotics: Science and Systems V, 2009, 2(4): 435-442.
[5]
KoideK, YokozukaM, OishiS, et al. Voxelized GICP for fast and accurate 3D point cloud registration[C]// 2021 IEEE International Conference on Robotics and Automation(ICRA). Xi’an: IEEE, 2021: 11054-11059.
[6]
DellenbachP, DeschaudJ, JacquetB, et al. CT-ICP: real-time elastic LiDAR odometry with loop closure[C]// 2022 International Conference on Robotics and Automation (ICRA). Philadelphia: ICRA, 2022: 5580-5586.
[7]
VizzoI, GuadagninoT, MerschB, et al. KISS-ICP: in defense of point-to-point ICP-simple, accurate, and robust registration if done the right way[J]. IEEE Robotics and Automation Letters, 2023, 8(2): 1029-1036.
[8]
ZhangJ, SinghS. Low-drift and real-time lidar odometry and mapping[J]. Auton Robot, 2017, 41(2): 401-406.
[9]
ShanT, EnglotB. LeGO-LOAM: lightweight and ground-optimized LiDAR odometry and mapping on variable terrain[C]// 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Madrid: IEEE, 2018: 4758-4765.
LiuZhenyu, HuiZeyu, GuoXu, et al. Tightly coupled simultaneous localization and mapping algorithm for LiDAR inertial measurement unit based on sliding window optimization[J]. Science Technology and Engineering, 2022, 22(21): 9167-9175.
JiangWenfei, ZhangXiaojun, WangJingang. Radar and camera fusion real-time positioning and mapping of underground garage[J]. Science Technology and Engineering, 2022, 22(6): 2334-2342.
[14]
ShanT, EnglotB, MeyersD, et al. LIO-SAM: tightly-coupled LiDAR inertial odometry via smoothing and mapping[C]// 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Las Vegas: IEEE, 2020: 5135-5142.
[15]
XuW, ZhangF. FAST-LIO: a fast, robust LiDAR-inertial odometry package by tightly-coupled iterated Kalman kilter[J]. IEEE Robotics and Automation Letters, 2022, 6(2): 3317-3324.
[16]
ChenK, NemiroffR, LopezB. Direct LiDAR-inertial odometry: lightweight LIO with continuous-time motion correction[C]// 2023 IEEE International Conference on Robotics and Automation (ICRA). London: IEEE, 2023: 3983-3989.
[17]
ChenZ J, XuY, YuanS H, et al. iG-LIO: an incremental GICP-based tightly-coupled LiDAR-inertial odometry[J]. IEEE Robotics and Automation Letters, 2024, 9(2): 1883-1890.
ChaiLiping, WangWeidong, LiChenyang, et al. Fusion SLAM method for LiDAR and IMU using surface features[J]. Laser & Optoelectronics Progress, 2025, 62(4): 197-204.
LiZhaoqiang, ZhangYue, XiongFuli, et al. Laser simultaneous location and mapping algorithm based on optimized ground segmentation and closed-loop detection[J]. Acta Optica Sinica, 2024, 44(20): 203-215.
TangHao, LiDong, WangCheng, et al. Global registration method for laser SLAM point clouds based on graph optimization[J]. Laser & Optoelectronics Progress, 2024, 61(10): 281-289.