计算机工程与应用 ›› 2022, Vol. 58 ›› Issue (18): 251-259.DOI: 10.3778/j.issn.1002-8331.2007-0489

• 图形图像处理 • 上一篇    下一篇

半结构化环境稀疏点云可通行区域检测方法

洪洋,袁夏,高飞,成诚,杨欢,陈耀忠,陆建峰   

  1. 1.南京理工大学,南京 210094
    2.海装上海局驻南京地区第四军事代表室,南京 210008
    3.北方信息控制研究院集团有限公司,南京 211100
  • 出版日期:2022-09-15 发布日期:2022-09-15

Traversable Area Detection for 3D Point Cloud in Semi-Structured Environment

HONG Yang, YUAN Xia, GAO Fei, CHENG Cheng, YANG Huan, CHEN Yaozhong, LU Jianfeng   

  1. 1.Nanjing University of Science and Technology, Nanjing 210094, China
    2.Nanjing Forth Military Representative Office of the Naval Equipment Department, Nanjing 210008, China
    3.North Information Control Research Academy Group Co., Ltd., Nanjing 211100, China
  • Online:2022-09-15 Published:2022-09-15

摘要: 传统的激光雷达可通行区域检测算法实时性好,但算法通常对点云数据的高度特征依赖较强,半结构化环境中斜坡等区域高度特征不稳定,易降低算法的性能。基于半结构化环境下三维稀疏点云数据,提出了一种基于直线特征的可通行区域实时检测算法,预处理原始点云数据,并从中速提取直线特征;对直线聚类和筛选,从而提取有效的直线特征;使用直线特征检测结构化障碍物和非结构化障碍物;标记出可通行区域,并构建环境地图。实验结果表明,在校园半结构化环境中,该算法可以较稳定地检测可通行区域,在检测路缘石等高度较低的障碍物时表现出良好的性能,并且适应斜坡、颠簸、负障碍等地面。

关键词: 三维稀疏点云, 直线特征提取, 半结构化环境, 可通行区域

Abstract: The traditional algorithms of detecting traversable area based on lidar perform well in real time. But the height feature of point cloud is one great influence on the algorithms, there are lots of unstable height features in semi-structured environment, it is easy to reduce the performance of the algorithms. Based on 3D sparse point cloud data, a method of detecting traversable area in semi-structured environment based on straight lines in real-time is constructed. Preprocessing original point cloud and the straight line features are quickly extracted. The effective lines are obtained from the straight lines by clustering and filtering. Then the structured obstacles and unstructured obstacles are detected by using the effective line features. The traversable area is marked, and the environment map is constructed. The experimental results show that in a semi-structured campus environment, the traversable area detected by the algorithm is more stably, also, very low obstacles can be detected steadily, such as curbs, and the algorithm can be used to detect slopes, bumps and negative obstacles.

Key words: 3D point cloud, line feature extraction, semi-structured environment, traversable area