基于双地图协同的动态障碍物点云检测与剔除算法

Dynamic Obstacle Point Cloud Detection and Removal Algorithm Based on Dual-map Collaboration

摘要

针对室内复杂场景中移动障碍物、镂空物体及设备倾斜等动态干扰因素导致激光 SLAM 系统点云特征退化、配准失效及地图畸变等问题，本文提出一种基于双地图协同的动态障碍物点云检测与剔除算法。首先，构建多分辨率栅格地图以优化相关性扫描匹配 (CSM) 算法的配准效率与精度；在此基础上，建立由静态基准地图与动态检测地图构成的双地图协同框架，通过双向约束机制实现对动态点云及虚假点云的精准判别与高效滤除。仿真与真实环境实验结果表明：该算法使 CSM 配准的匹配度稳定保持在 0.85 以上，有效规避了配准失效与地图畸变问题；动态障碍物识别率达到 98.75%，绝对轨迹误差 (ATE) 平均降低 3.675%。相较于传统栅格概率更新策略，本文算法从配准稳定性、动态环境感知能力及全局定位精度 3 个层面显著提升了激光 SLAM 系统的综合性能，能够构建无畸变的高可靠静态栅格地图，为室内动态场景下的自主定位与建图提供了有效的解决方案。

Abstract

To address the issues of point cloud feature uncertainty, registration failure, and map distortion in laser-based simultaneous localization and mapping (SLAM) caused by dynamic disturbances such as moving obstacles, hollow-structured objects, and equipment tilt in complex indoor scenes, a dynamic obstacle point cloud detection and removal algorithm based on dual-map collaboration was proposed. First, a multi-resolution grid map was constructed to optimize the registration efficiency and accuracy of the correlative scan matching (CSM) algorithm. On this basis, a dual-map system consisting of a static reference map and a local dynamic obstacle map was established, enabling accurate identification and efficient removal of dynamic and spurious point clouds through a bidirectional constraint mechanism. The experimental results from both simulation and real-world environments show that the proposed algorithm stabilizes the CSM registration score above 0.85, effectively avoiding registration failure and map distortion. The dynamic obstacle recognition rate reaches 98.75%, and the absolute trajectory error (ATE) is optimized by 3.68% on average. Compared with conventional grid-probability updating strategies, the proposed algorithm significantly improves the overall performance of laser SLAM systems in terms of registration stability, dynamic environmental perception, and global localization accuracy. A distortion-free and highly reliable static grid map can be constructed, providing an effective solution for autonomous localization and mapping in indoor dynamic scenes.

关键词

移动机器人 / 激光 SLAM / 动态点云去除 / 双地图协同 / 栅格地图构建 / 动态障碍物检测 / 相关性扫描匹配 (CSM)

Key words

mobile robot / laser slam / dynamic point cloud removal / dual-map collaboration / grid map construction / dynamic obstacle detection / correlative scan matching (CSM)

参考文献

[1] MUSLIM M T, SELAMAT H, ABURAYA A. YOSO−SLAM: a real—time object visual SLAM for dynamic scenes with semantic three—dimensional mapping[J/OL]. Arabian Journal for Science and Engineering, 2025−11−13. http://doi.org/10.1007/s13369-025-10840-4.

[2] 张硕, 李季轩, 宿玉康, 等. 基于动态点实时滤除与回环优化的 SLAM 方法[J]. 北京理工大学学报, 2026, 46(1): 47-60.

[3] ZHANG S, LI J X, SU Y K, et al. SLAM method based on dynamic point real—time filtering and loop closure optimization[J]. Beijing Institute of Launch Technology, 2026, 46(1): 47-60.

[4] SCHAUER J, NUECHTER A. The peopleremover—removing dynamic objects from 3D point cloud data by traversing a voxel occupancy grid[J]. IEEE Robotics and Automation Letters, 2018, 3(3): 1679-1686.

[5] LIM H, HWANG S, MYUNG H. ERASOR: Egocentric ratio of pseudo occupancy—based dynamic object removal for static 3D point cloud map building[J]. IEEE Robotics and Automation Letters, 2021, 6(2): 2272-2279.

[6] ZOU J, CHEN H, SHAO L, et al. DY−LIO: Tightly coupled lidar—inertial odometry for dynamic environments[J]. IEEE Sensors Journal, 2024, 24: 34756.

[7] 李擎, 林世杰, 贺晓东, 等. 基于动态点去除的激光雷达 SLAM 算法[J]. 工程科学学报, 2025, 47(10): 2070-2078.

[8] LI Q, LIN S J, HE X D, et al. LiDAR SLAM algotithm based on dynamic point removal[J]. Chinese Journal of Engineering, 2025, 47(10): 2070-2078.

[9] GHAHREMANI M, WILLIAMS K, CORKE F, et al. Direct and accurate feature extraction from 3D point clouds of plants using RANSAC[J]. Computers and Electronics in Agriculture, 2021, 187(13): 106240.

[10] SHAO Y, WANG B, TAN A. ADA−DPM: a neural descriptors—based adaptive noise point filtering strategy for SLAM[PP/OL]. arXiv (2025−10−20)[ 2025−03−06]. https://arxiv.org/abs/2506.18016.

[11] BESCÓS B, FÁCIL J M, CIVERA J, et al. DynaSLAM: tracking, mapping, and inpainting in dynamic scenes[J]. IEEE Robotics and Automation Letters, 2018, 3(4): 4076-4083.

[12] 马润辰, 马跃威, 安健硕, 等. 改进 PointNet++语义分割与三维重建融合的绿篱修剪智能决策框架[J/OL]. 北京林业大学学报, 2026−02−05. https://doi.org/10.12171/j.1000-1522.20260067.

[13] MA R C, MA Y W, AN J S, et al. A framework integrating improved PointNet++ segmentation and 3D reconstruction towards intelligent hedge trimming decisions[J/OL]. Journal of Beijing Forestry University, 2026−02−05. https://doi.org/10.12171/j.1000-1522.20260067.

[14] 幸荔芸, 李珊枝. 基于 Mask R−CNN 的激光雷达测量数据特征点识别[J]. 现代雷达, 2026, 48(1): 48-54.

[15] XING L Y, LI S Z. Feature point recognition of LiDAR measurement data based on Mask R−CNN[J]. Modern Radar, 2026, 48(1): 48-54.

[16] 郭致远, 刘瑞, 赵轩, 等. 动态场景下基于地面分割与回环优化的激光雷达定位与建图系统[J]. 计算机应用, 2025, 45(S1): 302-308.

[17] GUO Z Y, LIU R, ZHAO X, et al. Lidar SLAM system based on ground segmentation and loop closure optimization in dynamic environment[J]. Journal of Computer Applications, 2025, 45(S1): 302-308.

[18] LIAO P, CHEN L H, HU T, et al. IBR−SLAM: visual SLAM based on improved BiSeNet with RGB−D sensor[J]. Engineering Research Express, 2025, 7(3): 035229.

[19] 张爱武, 刘路路, 张希珍. 道路三维点云多特征卷积神经网络语义分割方法[J]. 中国激光, 2020, 47(4): 269-277.

[20] ZHANG A W, LIU L L, ZHANG X Z. Multi—feature 3D road point cloud semantic segmentation method based on convolutional neural network[J]. Chinese Journal of Lasers, 2020, 47(4): 269-277.

[21] GONG X G, MA J T, QIAN G P, et al. Aggregate evolution of asphalt mixture compaction process based on BiSeNet segmentation network: from mesoscopic to macroscopic[J]. International Journal of Pavement Engineering, 2025, 26(1): 2507125.

[22] 岳胜哲, 王正杰. 基于实例分割与光流的动态环境 SLAM[J]. 兵工学报, 2024, 45(1): 156-165.

[23] YUE S Z, WANG Z J. A SLAM in dynamic environment based on segmentation and optical flow[J]. Acta Armamentarii, 2024, 45(1): 156-165.

[24] 邓文轩, 党建武, 雍玖. 基于目标检测和点线特征关联的动态 SLAM 算法[J]. 激光与光电子学进展, 2025, 62(10): 1015007.

[25] DENG W X, DANG J W, YONG J. Dynamic SLAM algorithm based on object detection and point—line feature association[J]. Laser & Optoelectronics Progress, 2025, 62(10): 1015007.

[26] LUIS E, TONIX G, JOSÉ A, et al. Efficient deep learning—based M—PSK detection for OFDM V2V systems using MobileNetV3[J]. Algorithms, 2026, 19(3): 210.

[27] JIANG K T, WANG Y G, HE H Q, et al. Application of YOLOv7 and YOLOv8 transfer learning models in breast lesion classification and diagnosis.[J/OL]. Current Medical Imaging, 2026−02−02. https://doi.org/10.2174/0115734056422406251128144702.

[28] 陈仕豪, 何元烈, 刘铿. 融合G−ICP 与3D Gaussian splatting 的密集 SLAM 系统[J/OL]. 计算机应用研究, 2025−10−28. https://doi.org/10.19734/j.issn.1001-3695.2025.10.0418.

[29] CHEN S H, HE Y L, LIU Q. A aense SLAM system integrating G−ICP and 3D gaussian splatting[J/OL]. Application Research of Computers, 2025−10−28. https://doi.org/10.19734/j.issn.1001-3695.2025.10.0418.

[30] ZHANG Y, WANG X, LYU X, et al. Segment—based slam registration optimization algorithm combining NDT and PL−ICP[J]. Sensors., 2025, 25(23): 7175.

[31] LIU H, LUO S, LU J. Correlation scan matching algorithm based on multi—resolution auxiliary historical point cloud and lidar simultaneous localisation and mapping positioning application[J]. IET Image Processing., 2020, 14(14): 3596-3601.

[32] YUE H W, HOU L, SHEN P P, et al. A refined branch and bound algorithm for solving the sum of linear ratios programming[J]. Mathematics and Computers in Simulation, 2026, 247: 210-224.

[33] HESS W, KOHLER D, RAPP H, et al. Real—time loop closure in 2D Lidar SLAM[J]. IEEE, 2016: 1271-1278.

[34] 比特有灵实验室. ZIMA 激光扫地机器人仿真平台[EB/OL]. GitHub, (2025−02−27)[2026−03−06]. https://github.com/BitSoulLab/ZIMA.git.

[35] Bit Youling Laboratory. ZIMA laser sweeping robot simulation platform[EB/OL]. GitHub, (2025−02−27)[2026−03−06]. https://github.com/BitSoulLab/ZIMA.git.

基金资助

安徽省高等学校自然科学研究基金项目(2024AH051785)