汽车工程 ›› 2023, Vol. 45 ›› Issue (8): 1468-1478.doi: 10.19562/j.chinasae.qcgc.2023.08.017

所属专题: 智能网联汽车技术专题-感知&HMI&测评2023年

• • 上一篇    下一篇

非铺装道路凹凸不平特征语义分割方法研究

伍文广(),田双岳,张志勇,张斌   

  1. 长沙理工大学汽车与机械工程学院,长沙 410114
  • 收稿日期:2023-01-16 修回日期:2023-02-21 出版日期:2023-08-25 发布日期:2023-08-17
  • 通讯作者: 伍文广 E-mail:wwglq@csust.edu.cn
  • 基金资助:
    国家自然科学基金(52275086);湖南省自然科学基金(2021JJ30722);湖南省教育厅优青项目(21B0331)

Research on Semantic Segmentation of Uneven Features of Unpaved Road

Wenguang Wu(),Shuangyue Tian,Zhiyong Zhang,Bin Zhang   

  1. Changsha University of Science and Technology College,Changsha 410114
  • Received:2023-01-16 Revised:2023-02-21 Online:2023-08-25 Published:2023-08-17
  • Contact: Wenguang Wu E-mail:wwglq@csust.edu.cn

摘要:

非铺装道路凹凸不平特征参数复杂,增加了自动驾驶汽车提取有效信息以进行路径规划和决策控制的难度,精确的语义分割方法将有助于简化道路不平参数信息,提高特征识别精度,从而提高车辆自主行驶的安全性和舒适性。为此,本文提出了一种非铺装道路凹凸不平特征语义分割方法,实现对不同尺寸和高程差的凹凸不平特征分类。首先,引入高斯函数建立了非铺装道路表达模型,提出了凹凸不平特征自动标注方法并构建了仿真数据集,弥补了非铺装道路点云数据集的空白;然后,搭建了非铺装道路凹凸不平特征语义分割模型,基于Pointnet++的多层次特征提取结构,首次实现了对非铺装道路凹凸不平特征语义分割;最后,通过建立非铺装路面特征沙盘模型,运用本文提出的方法对实测数据进行验证。结果表明,本方法能准确对道路凹凸不平特征进行分类,且在不同点云数据密度、路面范围的数据中具有较好鲁棒性。

关键词: 非铺装道路, 凹凸不平特征, Pointnet++, 语义分割, 自动驾驶

Abstract:

The complexity of uneven feature parameters of unpaved roadincreases the difficulty of extracting effective information for path planning and decision control of autonomous vehicles. Accurate semantic segmentation method can help simplify road uneven parameter information, improve the accuracy and efficiency of feature recognition, thus improve the safety and comfort of autonomous driving of vehicles. Therefore, this paper proposes a semantic segmentation method for uneven features of unpaved road to classify uneven features of different dimensions and elevation differences. Firstly, the Gaussian function is introduced to establish the expression model of unpaved road, and the automatic labeling method of uneven features is proposed and the simulation data set is constructed, making up the gap of unpaved road point cloud data set. Then, a semantic segmentation model of unpaved road uneven features is built. Based on multi-level feature extraction structure of Pointnet++, the semantic segmentation of non-paved road uneven features is realized for the first time. Finally, the sand table model of unpaved road features is established and the proposed method is used for verification of the measured data. The results show that the proposed method can accurately classify the road uneven characteristics, and has good robustness in data of different point cloud data densities and road surface ranges.

Key words: unpaved road, uneven features, Pointnet++, semantic segmentation, automatic driving