基于地面特征的越野环境轻量化回环检测

doi:10.12382/bgxb.2024.0090

摘要/Abstract

摘要：

为解决传统回环检测算法依赖里程计精度与外部定位信息、消耗过多计算资源以及现有轻量化回环检测算法平移不变性差、难以适应越野场景下稀疏环境特征等问题,提升无人驾驶平台在卫星拒止条件下执行长时间、大范围行进任务的定位能力,提出一种利用激光雷达点云对地面特征进行描述的轻量化回环检测算法。不同于现有的利用深度学习从单帧或多帧点云中提取点云特征并构建全局描述子,所提算法使用快速的激光点云地面特征描述方式,实现单帧点云快速特征提取与全局一致的位置特征描述,将多帧激光点云地面特征聚合为子地图回环检测描述子;在临近帧之间依靠里程计位姿实现轻量化全局描述子的构建,不依赖先验位置信息进行全局描述子的匹配并实现回环检测;在越野环境下以机械式激光雷达、固态激光雷达对算法进行实车验证。研究结果表明,与已有的轻量化回环检测算法对比,验证了该算法在越野环境下回环检测高召回率、实时性好、占用资源少的优势。

关键词: 越野环境, 激光雷达, 地面特征, 回环检测

Abstract:

The traditional loop closure detection algorithm relies on the accuracy of odometer and the external global positioning information,which consumes too much computing resources,and the existing lightweight loop closure detection algorithm has poor translation invariance and difficulty in adapting to the sparse environmental characteristics in off-road road environment.In order to improve the positioning capability of unmanned platform in the condition of satellite rejection for a long time and a large range of tasks,a lightweight loop closure detection algorithm using light detection and ranging (LiDAR) point clouds to describe the ground feature is proposed.It is different from extracting the point cloud features from single or multi-frame point clouds by deep learning.And a global descriptor is constructed.The fast LiDAR point clouds ground feature description approach is used to achieve the fast feature extraction of single frame point cloud and the globally consistent position feature description,and the multi-frame LiDAR point clouds ground features are aggregated into the sub-map loop closure detection descriptors.A lightweight global descriptor is constructed by odometer pose between adjacent frames,and the global descriptors are matched and the loop closure detection is realized without prior position information.The proposed algorithm is verified by using the mechanical and solid-state LiDAR in off-road environment.Compared with the existing lightweight loop closure detection algorithms,the proposed algorithm has the advantages of high recall rate,good real-time performance and less resource consumption in the off-road environment.

Key words: off-road environment, light detection and ranging, ground feature, loop closure detection

张森杰, 龚建伟, 齐建永, 臧政, 胡秀中, 龚小杰, 熊光明. 基于地面特征的越野环境轻量化回环检测[J]. 兵工学报, 2025, 46(4): 240090-.

ZHANG Senjie, GONG Jianwei, QI Jianyong, ZANG Zheng, HU Xiuzhong, GONG Xiaojie, XIONG Guangming. Lightweight Loop Closure Detection of Off-road Environment Based on Ground Features[J]. Acta Armamentarii, 2025, 46(4): 240090-.

图/表 14

图1 无人平台定位系统框架

Fig.1 Framework of positioning system for unmanned platform

图2 基于激光点云地面特征的轻量化回环检测系统框架

Fig.2 Lightweight loop closure detection system framework based on ground features of LiDAR point clouds

图3 激光雷达地面点提取

Fig.3 Extraction of ground points from LiDAR points

图4 位姿修正后激光雷达单帧地面点提取

Fig.4 Extraction of ground points from LiDAR single frame points after the pose correction

图5 单帧激光点云与所属子地图的转换

Fig.5 Conversion of single frame LiDAR point clouds to submap

图6 子地图点云构建回环检测描述子

Fig.6 Loop closure detection descriptor made with point clouds in submap

表1 子地图回环检测描述子各通道内容

Table 1 Content of each channel of submap loop closure detection descriptor

通道	内容
B	地面点云回波强度,并从0~255归一化至0~155
G	3.1中计算的角度中大于5°的部分,并从5°~90°归一化至0~255
R	3.1中计算的角度中大于15°的部分,并从15°~90°归一化至0~255

图7 子地图回环检测描述子

Fig.7 Submap loop closure detection descriptor

表2 仿真实验与实车试验条件

Table 2 Simulation experiment and real vehicle test conditions

名称	编号	LiDAR数据来源	环境路况描述
测试Ⅰ	Ⅰ	Velodye_HDL_64	城市环境
测试Ⅱ	Ⅱ	RS_Helios_32	轨迹1
测试Ⅲ	Ⅲ-1	RS_Helios_32	轨迹4
	Ⅲ-2	RS_Helios_32	轨迹4反向
	Ⅲ-3	RS_LiDAR_M1	轨迹4
	Ⅲ-4	RS_LiDAR_M1	轨迹4反向
测试Ⅳ	Ⅳ-1	RS_Helios_32	轨迹2
测试Ⅳ	Ⅳ-2	RS_Helios_32	轨迹3

图8 实车测试轨迹1

Fig.8 Real test trajectory 1

图9 实车测试轨迹2、3、4

Fig.9 Real test trajectories 2,3 and 4

表3 本文算法耗时统计

Table 3 Time consumption statistics of the proposed algorithm ms

回环检测模块	LiDAR数据来源
回环检测模块	Velodye_HDL_64	RS_Helios_32	RS_LiDAR_M1
提取地面点	10.44	3.67	3.87
计算法向量	13.47	5.46	7.17
地面点插补至子地图	1.10	0.36	0.35
计算子地图描述子(每5s)	80	52.83	37.05
粗匹配	2.73	2.67	2.23
精匹配	→0	→0	→0
平均每帧总耗时	29.34	13.21	14.36

表4 开源算法耗时统计

Table 4 Time consumption statistics of open source algorithms ms

算法	回环检测模块	LiDAR数据来源
算法	回环检测模块	Velodye_HDL_64	RS_Helios_32	RS_LiDAR_M1
SC 算法	描述子构建	2.76	0.36	0.17
SC 算法	寻找回环	73.16	2.45	1.89
IRIS 算法	描述子构建	7.61	9.68	6.51
IRIS 算法	寻找回环	1913.51	1957.55	1945.35
M2DP 算法	描述子构建	1225.96	36.09	18.88
M2DP 算法	寻找回环	7.637	2.97	2.41

图10 准确度-召回率

Fig.10 Precision-recall

参考文献 23

[1]	陈慧岩, 张玉. 军用地面无人机动平台技术发展综述[J]. 兵工学报, 2014, 35(10):1696-1706. doi: 10.3969/j.issn.1000-1093.2014.10.026
	CHEN H Y, ZHANG Y. An overview of research on military unmanned ground vehicles[J]. Acta Armamentarii, 2014, 35(10):1696-1706. (in Chinese)
[2]	郑壮壮, 曹万科, 邹渊, 等. 动态环境下无人地面无人平台点云地图快速重定位方法[J]. 兵工学报, 2020, 41(8):1581-1589. doi: 10.3969/j.issn.1000-1093.2020.08.013
	ZHENG Z Z, CAO W K, ZOU Y, et al. Rapid localization of unmanned ground vehicles in dynamic environment using point cloud maps[J]. Acta Armamentarii, 2020, 41(8):1581-1589. (in Chinese) doi: 10.3969/j.issn.1000-1093.2020.08.013
[3]	JI K J, CHEN H Y, DI H J, et al.CPFG-SLAM:a robust Simultaneous localization and mapping based on LiDAR in off-road environment[C]//Proceedings of 2018 IEEE Intelligent Vehicles Symposium (Ⅳ).Changshu,China:IEEE,2018:650-655.
[4]	张福斌, 张炳烁, 杨玉帅. 基于单目/IMU/里程计融合的SLAM算法[J]. 兵工学报, 2022, 43(11):2810-2817.
	ZHANG F B, ZHANG B S, YANG Y S. SLAM algorithm based on monocular/IMU/odometer fusion[J]. Acta Armamentarii, 2022, 43(11):2810-2817. (in Chinese) doi: 10.12382/bgxb.2022.0240
[5]	CAMPOS C, ELVIRA R, RODRIGUEZ J J G, 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.
[6]	QIN T, LI P, SHEN S J. VINS-Mono:A robust and versatile monocular visual-inertial state estimator[J]. IEEE Transactions on Robotics, 2018, 34(4):1004-1020.
[7]	CUI Y, CHEN X, ZHANG Y L, et al. BoW3D:Bag of words for real-time loop closing in 3D LiDAR SLAM[J]. IEEE Robotics and Automation Letters, 2023, 8(5):2828-2835.
[8]	HESS W, KOHLER D, RAPP H, et al. real-time loop closure in 2D liDAR SLAM[C]// Proceedings of 2016 IEEE International Conference on Intelligent Robots and Systems.Stockholm, Sweden:IEEE,2016:1271-1278.
[9]	SHAN T, ENGLOT B, MEYERS D, et al. LIO-SAM:tightly-coupled LiDAR inertial odometry via Smoothing and Mapping[C]// Proceedings of 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems.Las Vegas,NV, US: IEEE,2020:5135-5142.
[10]	HE L, WANG X L, ZHANG H, et al. M2DP:a novel 3D point cloud descriptor and its application in loop closure detection[C]// Proceedings of 2016 IEEE/RSJ International Conference on Intelligent Robots and Systems.Daejeon, Korea(South):IEEE,2016:231-237.
[11]	KIM G, KIM A. Scan context:egocentric spatial descriptor for place recognition within 3D point cloud map[C]// Proceedings of 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems.Madrid, Spain:IEEE,2018:4802-4809.
[12]	WANG Y, SUN Z Z, XU C Z, et al. LiDAR Iris for loop-closure detection[C]// Proceedings of 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems.Las Vegas,NV, US: IEEE,2020:5769-5775.
[13]	蒋旭东. 基于多传感器融合的紧耦合SLAM算法的设计与实现[D]. 长春: 吉林大学, 2023.
	JIANG X D. Design and implementation of tightly coupled SLAM algorithm based on multi-sensor[D]. Changchun: Jilin University, 2023. (in Chinese)
[14]	张凯钧. 基于复杂场景的高精度地图生成算法研究[D]. 长春: 吉林大学, 2023.
	ZHANG K J. Research on high-precision map generation algorithm based on complex scene[D]. Changchun: Jilin University, 2023. (in Chinese)
[15]	JOO H J, KIM J. RID:LiDAR range image descriptor for fast and efficient loop closure detection in indistinguishable environments[C]// Proceedings of 2013 IEEE International Conference on Metaverse Computing,Networking and Applications.Kyoto, Japan:IEEE,2023:742-748.
[16]	刘忠泽, 陈慧岩, 崔星, 等. 无人平台越野环境下同步定位与地图创建[J]. 兵工学报, 2019, 40(12):2399-2406. doi: 10.3969/j.issn.1000-1093.2019.12.002
	LIU Z Z, CHEN H Y, CUI X, et al. Real-time LiDAR SLAM in off-road environment for UGV[J]. Acta Armamentarii, 2019, 40(12):2399-2406. (in Chinese) doi: 10.3969/j.issn.1000-1093.2019.12.002
[17]	高馨雨. 起伏路环境激光SLAM算法设计与实现[D]. 长春: 吉林大学, 2023.
	GAO X Y. Design and implementation of simultaneous localization and mapping system for undulating road environments[D]. Changchun: Jilin University, 2023. (in Chinese)
[18]	DUBE R, DUGAS D, STUMM E, et al. SegMatch:segment based place recognition in 3D point clouds[C]// Proceedings of 2017 IEEE International Conference on Robotics and Automation. Singapore:IEEE,2017:5266-5272.
[19]	CAO D Y, LYU Z G, LIU Z, et al. Loop closure detection based on geometry of semantic point cloud[C]// Proceedings of 2023 IEEE 18th Conference on Industrial Electronics and Applications.Ningbo, China: IEEE,2023:1668-1673.
[20]	XU M, LIN S Q, WANG J K, et al. A LiDAR SLAM system with geometry feature group-based stable feature selection and three-stage loop closure optimization[J]. IEEE Transactions on Instrumentation and Measurement, 2023,72:8504810.
[21]	MA J Y, ZHANG J, XU J T, et al. OverlapTransformer:an efficient and yaw-angle-invariant transformer network for LiDAR-based place recognition[J]. IEEE Robotics and Automation Letters, 2022, 7(3):6958-6965.
[22]	LI L, KONG X, ZHAO X R, et al. RINet:efficient 3D Lidar-based place recognition using rotation invariant neural network[J]. IEEE Robotics and Automation Letters, 2022, 7(2):4321-4328.
[23]	KOMOROWSKI J. MinkLoc3D:point cloud based large-scale place recognition[C]// Proceedings of 2021 IEEE Winter Conference on Applications of Computer Vision.Waikoloa,HI, US: IEEE,2021:1789-1798.