汽车工程 ›› 2025, Vol. 47 ›› Issue (10): 1872-1884.doi: 10.19562/j.chinasae.qcgc.2025.10.003

• • 上一篇    

基于改进PointNet++的非铺装路面凹凸不平识别和行车风险建模

伍文广1(),邱松江1,胡林1,张晓强2,邱增华3,汪嘉凯1   

  1. 1.长沙理工大学机械与运载工程学院,长沙 410114
    2.湘电重型装备有限公司,湘潭 411101
    3.湘电集团有限公司,湘潭 411101
  • 收稿日期:2025-03-28 修回日期:2025-05-05 出版日期:2025-10-25 发布日期:2025-10-20
  • 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(52325211);国家自然科学基金(52275086);湖南省自然科学基金(2025JJ50309);湖南省自然科学基金(2021JJ30722);湖南省普通高等学校科技创新团队(2024RC1029)

Unpaved Road Unevenness Recognition and Driving Risk Model Construction Based on the Improved Pointnet++

Wenguang Wu1(),Songjiang Qiu1,Lin Hu1,Xiaoqiang Zhang2,Zenghua Qiu3,Jiakai Wang1   

  1. 1.College of Mechanical and Vehicle Engineering,Changsha University of Science and Technology,Changsha 410114
    2.Xiangtan Electric Manufacturing Group Heavy-Duty Equipment Co. ,Ltd. ,Xiangtan 411101
    3.Xiangtan Electric Manufacturing Group Co. ,Ltd. ,Xiangtan 411101
  • Received:2025-03-28 Revised:2025-05-05 Online:2025-10-25 Published:2025-10-20
  • Contact: Wenguang Wu E-mail:wwglq@csust.edu.cn

摘要:

非铺装路面的凹坑和凸起对车辆行驶安全性有重要影响,但是这些不平特征既不能视作障碍物,也不能当作路面粗糙度,给非铺装道路下车辆的路径规划和行驶安全性评估带来挑战。为此,本文根据路面凹凸不平的高度/深度、面积和形状等对车辆行驶安全性的影响,提出一种基于激光雷达点云的车辆行驶风险模型构建方法。首先,针对非铺装道路凹坑凸起不平特征数据特点,提出改进PointNet++模型,对凹凸不平进行识别并提升点云语义分割性能;然后,根据语义分割结果和感兴趣点云类别定义,采用3种聚类算法对点云数据进行聚类对比,确定采用DBSCAN算法实现非铺装道路下的最高效率技术路线;最后,根据不同凹坑和凸起对车辆行驶安全性的影响,引入地形类别系数和地形密度风险,基于二维高斯函数构建了凹凸不平对车辆的安全风险模型,实现对非铺装道路场景驾驶风险的参数化表达。通过实验和仿真数据对比,结果表明,本文改进的PointNet++语义分割模型平均联合交集提高了5.7%,准确率和召回率分别提升5.3%和6.4%,实现了非铺装路面凹凸不平高精度识别和行驶风险建模,在不同点云密度、路面范围和道路场景下均具有较高的鲁棒性。

关键词: 非铺装道路, 凹凸不平, 风险模型, 环境感知

Abstract:

Potholes and bulges on unpaved road have a significant impact on vehicle driving safety. However, these uneven features cannot be simply regarded as obstacles or surface roughness, posing challenges for vehicle path planning and driving safety assessment in unpaved road environment. To address this, a LiDAR point cloud-based vehicle risk modeling method is proposed, considering the effect of surface unevenness, such as height/depth, area, and shape, on driving safety. Firstly, for the characteristics of pothole and bump features in unpaved road point cloud data, an improved PointNet++ model is proposed to enhance semantic segmentation performance and accurately identify uneven structures. Then, based on the semantic segmentation results and the defined categories of interest, three clustering algorithms are compared, and DBSCAN is identified as the most efficient method for clustering uneven features in unpaved road. Finally, to reflect the impact of various potholes and bumps on vehicle safety, terrain category coefficients and terrain density risks are introduced. A two-dimensional Gaussian function is used to construct a risk model for uneven terrain, enabling parametric representation of driving risks in unpaved road scenarios. The experimental and simulation results show that the improved PointNet++ model increases mIoU by 5.7%, and improves accuracy and recall by 5.3% and 6.4%, respectively. The proposed method achieves high-precision recognition of unpaved road unevenness and driving risk modeling, exhibiting strong robustness across different point cloud densities, road coverage, and road scenarios.

Key words: unpaved road, unevenness, risk model, environmental perception