北京航空航天大学学报 ›› 2021, Vol. 47 ›› Issue (2): 306-314.doi: 10.13700/j.bh.1001-5965.2020.0212

• 论文 • 上一篇    下一篇

基于多约束因子图优化的无人车定位与建图方法

牛国臣, 王瑜   

  1. 中国民航大学 机器人研究所, 天津 300300
  • 收稿日期:2020-05-25 发布日期:2021-03-08
  • 通讯作者: 牛国臣 E-mail:niu_guochen@139.com
  • 作者简介:牛国臣,男,博士,副教授,硕士生导师。主要研究方向:智能机器人、机场智能与自动化系统;王瑜,女,硕士研究生。主要研究方向:激光SLAM、自主导航系统。
  • 基金资助:
    中央高校基本科研业务费专项资金(3122019043);天津市科技计划(17ZXHLGX00120)

Unmanned vehicle positioning and mapping method based on multi-constraint factor graph optimization

NIU Guochen, WANG Yu   

  1. Institute of Robotics, Civil Aviation University of China, Tianjin 300300, China
  • Received:2020-05-25 Published:2021-03-08

摘要: 针对目前在特定场景下应用的低速无人车定位系统极度依赖全球导航卫星系统(GNSS),存在定位精度不高、漂移误差大、受环境影响严重等问题,提出一种低成本、高精度的无人车定位与建图方法。该方法基于三维激光定位与建图(SLAM)技术。首先,使用点云主成分分析(PCA)实现基于特征匹配的激光里程计;其次,将GNSS位置信息、点云分割聚类得到的地平面和点云聚类特征作为位姿约束分别加入图优化框架,消除激光里程计的累积误差;最后,得到最优位姿和大规模场景的点云地图,以实现无人车的自主定位导航。利用包含大型户外城市街道环境的KITTI数据集对所提出的SLAM算法进行了评估,结果表明:系统在3km运动距离情况下定位偏差可控制在1.5 m以下,在局部精度和全局一致性方面均优于其他里程计系统,为无人车的定位提供了新思路。

关键词: 图优化, 三维激光定位与建图(SLAM), 点云分割, 主成分分析(PCA), 无人车

Abstract: Aimed at the problem that the current low-speed positioning system of unmanned vehicle extremely relies on the Global Navigation Satellite System (GNSS), which has low positioning accuracy, large drift error and serious environmental impact, a low-cost and high-precision positioning and mapping method is proposed. This method is based on the three-dimensional laser Simultaneous Localization and Mapping(SLAM) technology. First, the point cloud Principal Component Analysis (PCA) is used to implement laser odometry based on feature matching. Then, the GNSS location information, ground plane and clustering feature of point cloud obtained by point cloud segmentation and clustering are added to the graph optimization framework as pose constraints, and the cumulative error of the laser odometry is eliminated. Finally, an optimal pose and large-scale scenes point cloud map is obtained to achieve the unmanned vehicles' position navigation. The proposed SLAM algorithm is evaluated using the KITTI dataset containing large outdoor urban street environments. The results show that the positioning deviation of this system can be controlled below 1.5 m at a movement distance of 3 km, and both in terms of local accuracy and global consistency, it is superior to other odometry systems and provides new ideas for the positioning of unmanned vehicles.

Key words: graph optimization, three-dimensional laser Simultaneous Localization and Mapping (SLAM), point cloud segmentation, Principal Component Analysis (PCA), unmanned vehicle

中图分类号: 


版权所有 © 《北京航空航天大学学报》编辑部
通讯地址:北京市海淀区学院路37号 北京航空航天大学学报编辑部 邮编:100191 E-mail:jbuaa@buaa.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发