胡肖建,李少达,谭骏祥,等.一种基于位姿图的激光SLAM点云优化方法[J].河南理工大学学报(自然科学版)，doi:10.16186/j.cnki.1673-9787.2025010030.

HU X J, LI S D, TAN J X, et al． An optimization method for laser SLAM point clouds based on pose graph[J]. Journal of Henan Polytechnic University( Natural Science) ， doi: 10.16186/j.cnki.1673-9787.2025010030.

一种基于位姿图的激光SLAM点云优化方法(网络首发)

胡肖建1，李少达2，谭俊祥1，胡古月1，刘涛1，杨小军2

1.成都理工大学 地球与行星科学学院，四川 成都 610059；2.中国电力工程顾问集团西南电力设计院有限公司，四川 成都 610056

摘要: 基于同步定位和制图（simultaneous localization and mapping, SLAM）技术的激光扫描系统因成本低和效率高等优势在测绘领域得到了广泛应用，但大量研究表明SLAM实时建图误差会随着扫描进行累计误差增大，导致重访点云重影现象严重。目的 本文旨在提出一种基于位姿图的SLAM点云数据优化方法，提高点云数据局部一致性与整体精度。 方法 该方法首先设计了一种联合时间与空间动态变化的点云分段策略；随后，构建了结合位姿节点距离与点云边缘点全局特征相似度的回环检测算法，以查找具有约束关系的位姿节点，并通过基于正态变换分布粗配准与“点-切平面”迭代最近邻精配准得出位姿节点间约束关系。最后，在局部与整体两个层次分别构建位姿图对轨迹进行优化，借助优化后的轨迹修正点云。结果 为验证方法的有效性，使用4组车载激光SLAM数据进行实验，经过优化后位姿估计精度分别提高81.4%，93.4%，81.2%，66.6%。4组点云同名点对间的均方根误差由原来的268.4，169.3，138.5，89.6 cm降为25.6， 31.2，15.7，11.3 cm。优化后同一地物的不同扫描帧点云间漂移现象减弱，点云同名点对间的均方根误差减少约87%，点云内部不一致性现象得到有效消除。结论 本文方法能够解决长距离扫描的漂移误差问题，为高精度点云获取提供重要的技术支撑。

关键词:同步定位与制图；图优化；点云配准；回环检测；点云修正

doi: 10.16186/j.cnki.1673-9787.2025010030.

基金项目:国家自然科学基金资助项目（42401425）；海南省重点研发项目（ZDYF2025XDNY113）

收稿日期：2025-01-22

修回日期：2025-04-27

网络首发日期：2025-12-12

An optimization method for laser SLAM point clouds based on pose graph(Online)

Hu Xiaojian 1, Li Shaoda 1, Tan Junxiang 1, HU Guyue1, Liu Tao1, Yang Xiaojun 2

1.School of Earth and Planetary Sciences, Chengdu University of Technology, Chengdu 610059, Sichuan, China；2.Southwest Electric Power Design Institute Co.,Ltd. of China Power Engineering Consulting Group，Chengdu 610056, Sichuan, China

Abstract: The Laser scanning system based on Simultaneous Localization and Mapping (SLAM) has been widely used in the field of mapping. It is favored for their low cost and high efficiency. However, numerous studies have shown that SLAM real-time mapping errors accumulate over time during scanning, which leads to severe ghosting effects in revisited point clouds. Objectives This paper aims to propose a SLAM point clouds optimization method based on pose graph to improve the local consistency and overall accuracy of point clouds. Methods The proposed method first designs a point clouds segmentation strategy that jointly considers temporal and spatial dynamics. Then, a loop closure detection algorithm is constructed, which integrates the pose node distance and the global feature similarity of point clouds edge points to find pose nodes with constraints. The constraints between pose nodes are derived through coarse registration based on Normal Distributions Transform(NDT) and fine registration using "point-to-plane" Iterative Closest Point(ICP). Finally, pose graph optimization is performed at both local and global levels, and the corrected trajectory is used to refine the point clouds. Results To validate the effectiveness of the proposed method, experiments were conducted using four sets of vehicle-mounted laser SLAM data. After optimization, the pose estimation accuracy was improved by 81.4%, 93.4%, 81.2%, and 66.6%, respectively. The root mean square error (RMSE) of corresponding point pairs in the four point clouds was reduced from 268.4, 169.3, 138.5, and 89.6 cm to 25.6, 31.2, 15.7, and 11.3 cm, respectively. The experimental results showed that after optimization, the drift phenomenon between different scan frame point clouds of the same object was reduced. The root mean square error between corresponding points decreased by approximately 87%. The internal inconsistency of the point clouds was effectively eliminated. Conclusions This method can address the drift error problem in long-distance scanning and provides significant technical support for high-precision point clouds acquisition.

Key words: simultaneous localization and mapping; graph optimization; point clouds registration; loop closure detection; point clouds correction