引用本文: 黄思源，刘利民，董健，等. 车载激光雷达点云数据地面滤波算法综述[J]. 光电工程，2020， 47 (12)：190688 . doi: 10.12086/oee.2020.190688 Citation: Huang S Y, Liu L M, Dong J, et al. Review of ground filtering algorithms for vehicle LiDAR scans point cloud data[J]. Opto-Electron Eng , 2020, 47 (12): 190688 . doi: 10.12086/oee.2020.190688

Abstract

LiDAR plays an important role in the field of unmanned driving. Ground filtering is the key technology to separate and extract the ground information from the point cloud data acquired by LiDAR. Firstly, the development and classification of vehicle LiDAR scans (VLS) are introduced, and the advantages and disadvantages of all kinds of VLS are discussed. Then, the development of VLS ground filtering algorithm is studied and classified. The evaluation methods and standards of ground filtering accuracy are described, and three typical algorithms are compared and analyzed. Finally, the shortcomings of current VLS and its ground filtering algorithms are summarized, and the future development trend is prospected.

Key words: LiDAR ground filtering intelligent driving accuracy evaluation prospect

Overview

Overview: LiDAR plays an important role in the field of unmanned driving. Ground filtering is the key technology to separate and extract the ground information according to the point cloud data acquired by LiDAR. First of all, this paper briefly describes the landmark events that vehicle LiDAR scans (VLS) established its position in the field of unmanned driving. According to the classification of mechanical, mixed solid and solid LiDAR, the working principle of each type of VLS is described, and the advantages and disadvantages of each type of VLS are discussed and compared. Secondly, the development of VLS ground filtering algorithms is studied. And the existing algorithms are sorted according to the processing methods of point cloud data. The ground filtering algorithm is divided into four categories: the ground filtering algorithm based on space division, the ground filtering algorithm based on scan lines, the ground filtering algorithm based on local characteristics, and the ground filtering algorithm based on additional information. According to the principles and filtering results of different algorithms, their characteristics, advantages and disadvantages are described. In addition to the above filtering algorithms, some ground filtering algorithms are also introduced. However, the adaptability of these algorithms to VLS point cloud data needs to be further improved. The common evaluation methods and standards of ground filtering accuracy are described to effectively evaluate the filtering results of various algorithms in different situations. There are three evaluation methods of filtering results: the manual calibration method, the visual inspection method, and the random sampling method. Furthermore, there are three evaluation standards for filtering accuracy: the cross table method, the Kappa coefficient method, and the algorithm time/space complexity. In order to show the characteristics of various algorithms, typical algorithms are selected for comparison from the ground filtering algorithm based on spatial division, the ground filtering algorithm based on scan lines and the ground filtering algorithm based on local characteristics. By changing the selected value of parameters, several groups of tests are carried out for each algorithm. The filtering results are arranged in ascending order according to Kappa coefficient, and the influence of parameter changes on the results is analyzed. The accuracy evaluation criteria are used to compare and analyze the optimal filtering results. Finally, the shortcomings of existing VLS ground filtering algorithms are summarized, and the development trend of VLS and VLS ground filtering algorithms is prospected. With the development of the computer technology and machine learning technology, filtering algorithms will be more intelligent and efficient.

大扫描视场和高扫描效率，可承受的激光功率高 机械结构复杂，设备难以小型化，行车环境下磨损严重，使用寿命短，价格高昂 Velodyne公司(美国)
Quanergy公司(德国)
上海禾赛光电
深圳速腾聚创 混合固态
激光雷达 通过MEMS振镜旋转完成激光扫描 实现了一定程度的小型化，响应速度较快 接收光路复杂，使用寿命短，扫描受限于振镜的偏转范围 Msotek公司(韩国)
Innoviz公司(以色列) 光学相控阵型
激光雷达 通过控制合成光束的指向完成激光扫描 无惯性器件，精确稳定，方向可任意控制 需要消除旁瓣的影响，难以实现水平360°扫描 Quanergy公司(美国)
Blackmore公司(美国) 闪光型
激光雷达 采用单脉冲直接向各个方向漫射，利用飞行时间成像 只要一次快闪便能照亮整个场景，避免运动畸变 探测精度随距离增加明显降低，视场角受限 亚德诺半导体公司
(美国) 刘志青, 黄沈华, 马琪, 等.基于混合最小二乘与总体最小二乘的激光雷达滤波算法[J].测绘与空间地理信息, 2019, 42 (2): 30-33. https://www.cnki.com.cn/Article/CJFDTOTAL-DBCH201902009.htm Liu Z Q, Huang S H, Ma Q, et al . LiDAR filtering algorithm based on mixed least squares and total least squares[J]. Geomatics & Spatial Information Technology , 2019, 42 (2): 30-33. https://www.cnki.com.cn/Article/CJFDTOTAL-DBCH201902009.htm 邱纯鑫.激光雷达与自动驾驶的产业化之路[J].人工智能, 2018(6): 37-47. https://www.cnki.com.cn/Article/CJFDTOTAL-DKJS201806006.htm Qiu C X. Lidar and the industrialization of automatic driving[J]. Artificial Intelligence , 2018(6): 37-47. https://www.cnki.com.cn/Article/CJFDTOTAL-DKJS201806006.htm 陈晓冬, 张佳琛, 庞伟凇, 等.智能驾驶车载激光雷达关键技术与应用算法[J].光电工程, 2019, 46 (7): 190182. doi: 10.12086/oee.2019.190182 Chen X D, Zhang J C, Pang W S, et al . Key technology and application algorithm of intelligent driving vehicle LiDAR[J]. Opto-Electronic Engineering , 2019, 46 (7): 190182. doi: 10.12086/oee.2019.190182 Hata A Y, Wolf D F. Feature detection for vehicle localization in urban environments using a multilayer LIDAR[J]. IEEE Transactions on Intelligent Transportation Systems , 2016, 17 (2): 420-429. http://ieeexplore.ieee.org/document/7279128/ Yin H L, Yang X H, He C. Spherical coordinates based methods of ground extraction and objects segmentation using 3-D LiDAR sensor[J]. IEEE Intelligent Transportation Systems Magazine , 2016, 8 (1): 61-68. http://ieeexplore.ieee.org/document/7384616/ 张名芳, 付锐, 郭应时, 等.基于三维不规则点云的地面分割算法[J].吉林大学学报(工学版), 2017, 47 (5): 1387-1394. https://www.cnki.com.cn/Article/CJFDTOTAL-JLGY201705009.htm Zhang M F, Fu R, Guo Y S, et al . Road segmentation method based on irregular three dimensional point cloud[J]. Journal of Jilin University (Engineering and Technology Edition) , 2017, 47 (5): 1387-1394. https://www.cnki.com.cn/Article/CJFDTOTAL-JLGY201705009.htm 董敏, 陈铁桩, 杨浩.基于Mesh的地面激光点云分离方法研究[J].计算机工程, 2019, 45 (6): 32-36, 44. https://www.cnki.com.cn/Article/CJFDTOTAL-JSJC201906006.htm Dong M, Chen T Z, Yang H. Research on separation method of ground laser point cloud based on mesh[J]. Computer Engineering , 2019, 45 (6): 32-36, 44. https://www.cnki.com.cn/Article/CJFDTOTAL-JSJC201906006.htm 苏本跃, 马金宇, 彭玉升, 等.基于K-means聚类的RGBD点云去噪和精简算法[J].系统仿真学报, 2016, 28 (10): 2329-2334, 2341. https://www.cnki.com.cn/Article/CJFDTOTAL-XTFZ201610006.htm Su B Y, Ma J Y, Peng Y S, et al . Algorithm for RGBD point cloud denoising and simplification based on K-means clustering[J]. Journal of System Simulation , 2016, 28 (10): 2329-2334, 2341. https://www.cnki.com.cn/Article/CJFDTOTAL-XTFZ201610006.htm Biosca J M, Lerma J L. Unsupervised robust planar segmentation of terrestrial laser scanner point clouds based on fuzzy clustering methods[J]. ISPRS Journal of Photogrammetry and Remote Sensing , 2008, 63 (1): 84-98. http://www.sciencedirect.com/science/article/pii/s0924271607000809 Zhou W Q. An object-based approach for urban land cover classification: integrating LiDAR height and intensity data[J]. IEEE Geoscience and Remote Sensing Letters , 2013, 10 (4): 928-931. http://ieeexplore.ieee.org/document/6497495/ Franceschi M, Teza G, Preto N, et al . Discrimination between marls and limestones using intensity data from terrestrial laser scanner[J]. ISPRS Journal of Photogrammetry and Remote Sensing , 2009, 64 (6): 522-528. http://www.sciencedirect.com/science/article/pii/S0924271609000446 Pirotti F, Guarnieri A, Vettore A. Ground filtering and vegetation mapping using multi-return terrestrial laser scanning[J]. ISPRS Journal of Photogrammetry and Remote Sensing , 2013, 76 : 56-63. http://www.sciencedirect.com/science/article/pii/S0924271612001505 Boyko A, Funkhouser T. Extracting roads from dense point clouds in large scale urban environment[J]. ISPRS Journal of Photogrammetry and Remote Sensing , 2011, 66 (6): S2-S12. http://www.sciencedirect.com/science/article/pii/S0924271611001067 Song H, Choi W, Kim H. Robust vision-based relative-localization approach using an RGB-depth camera and LiDAR sensor fusion[J]. IEEE Transactions on Industrial Electronics , 2016, 63 (6): 3725-3736. http://ieeexplore.ieee.org/document/7390258 Lichti D D. Spectral filtering and classification of terrestrial laser scanner point clouds[J]. The Photogrammetric Record , 2005, 20 (111): 218-240. http://onlinelibrary.wiley.com/doi/10.1111/j.1477-9730.2005.00321.x Thrun S. Learning occupancy grid maps with forward sensor models[J]. Autonomous Robots , 2003, 15 (2): 111-127. http://dl.acm.org/citation.cfm?id=940152.940193 Kammel S, Ziegler J, Pitzer B, et al . Team AnnieWAY's autonomous system for the 2007 DARPA urban challenge[J]. Journal of Field Robotics , 2008, 25 (9): 615-639. http://dl.acm.org/citation.cfm?id=1405648 Zhang K Q, Chen S C, Whitman D, et al . A progressive morphological filter for removing nonground measurements from airborne LIDAR data[J]. IEEE Transactions on Geoscience and Remote Sensing , 2003, 41 (4): 872-882. http://ieeexplore.ieee.org/document/1202973 黄作维, 刘峰, 胡光伟.基于多尺度虚拟格网的LiDAR点云数据滤波改进方法[J].光学学报, 2017, 37 (8): 0828004. https://www.cnki.com.cn/Article/CJFDTOTAL-GXXB201708042.htm Huang Z W, Liu F, Hu G W. Improved method for LiDAR point cloud data filtering based on hierarchical pseudo-grid[J]. Acta Optica Sinica , 2017, 37 (8): 0828004. https://www.cnki.com.cn/Article/CJFDTOTAL-GXXB201708042.htm Cohen J. A coefficient of agreement for nominal scales[J]. Educational and Psychological Measurement , 1960, 20 (1): 37-46. http://epm.sagepub.com/content/74/1/116/F1.expansion.html Geiger A, Lenz P, Stiller C, et al . Vision meets robotics: the KITTI dataset[J]. The International Journal of Robotics Research , 2013, 32 (11): 1231-1237. http://imaiai.oxfordjournals.org/external-ref?access_num=10.1177/0278364913491297&link_type=DOI Li J, Mei X, Prokhorov D, et al . Deep neural network for structural prediction and lane detection in traffic scene[J]. IEEE Transactions on Neural Networks and Learning Systems , 2017, 28 (3): 690-703. http://europepmc.org/abstract/med/26890928 Figure 1.

Space division methods of point cloud. (a) Local point cloud; (b) Gridding; (c) Voxel partition

Figure 2.

Change of scanning lines. (a) Scanning scene simulation of VLS; (b) Horizontal projection of scanning lines

Figure 3.

The scene of point cloud data

Figure 4.

Calibration results

Figure 5.

Filtering results of three typical ground filtering algorithms. (a) Results of growing up algorithm; (b) Results of adjacent points; (c) Results of slope-regional growth; (d) Point cloud of growing up algorithm; (e) Point cloud of adjacent points; (f) Point cloud of slope-regional growth