汽车工程 ›› 2024, Vol. 46 ›› Issue (10): 1842-1852.doi: 10.19562/j.chinasae.qcgc.2024.10.011

• • 上一篇    下一篇

基于点云反射特性的前方道路附着系数估计方法研究

胡宏宇,唐明弘,高菲,鲍明喜,高镇海   

  1. 吉林大学,汽车底盘集成与仿生全国重点实验室,长春 130022
  • 收稿日期:2024-05-11 修回日期:2024-06-24 出版日期:2024-10-25 发布日期:2024-10-21
  • 通讯作者: 高镇海
  • 基金资助:
    吉林省重大科技专项(20220301009GX);重载车辆模块化可扩展构型设计研究项目(52394261);吉林省科技发展计划项目(202302013)

Research on the Estimation Method of Road Friction Coefficient Ahead Based on Point Cloud Reflection Properties

Hongyu Hu,Minghong Tang,Fei Gao,Mingxi Bao,Zhenhai Gao   

  1. Jilin University,National Key Laboratory of Automotive Chassis Integration and Bionics,Changchun 130022
  • Received:2024-05-11 Revised:2024-06-24 Online:2024-10-25 Published:2024-10-21
  • Contact: Zhenhai Gao

摘要:

路面附着系数是影响自动驾驶系统决策控制策略的重要因素。为实现对道路附着系数前瞻性的高精度感知,本文基于车载激光雷达设计了一种新的路面附着系数估计方法。首先采集了干燥柏油路面、混凝土路面、湿滑柏油路面、结冰路面和积雪路面构建道路数据集;基于使用布料模拟滤波和RANSAC算法进行了道路点云提取、基于高斯滤波去除反射率异常噪点;根据点云反射率随距离和入射角变化的规律将路面划分为不同区域分别提取特征;基于深度神经网络构建了道路识别模型,并基于采集数据集进行了训练,最后基于路面材质和峰值附着系数的统计经验确定了前方道路的附着系数。测试结果表明,本文提出的算法道路类型辨识精度超过99.3%,算法平均运行周期55 ms,可实现实时高精度的路面峰值附着系数估计。

关键词: 路面附着系数, 激光雷达点云, 布料模拟滤波, RANSAC, 深度神经网络, 高斯滤波, 路面类型识别

Abstract:

The road friction coefficient is a significant factor that impacts the decision-making control strategy of the autonomous driving system. To achieve prospective and high-precision perception of the road friction coefficient, a novel estimation method for road friction coefficient based on the LiDAR equipped in vehicles is proposed in this paper. Firstly, a road dataset is constructed by collecting data from dry asphalt, concrete, wet asphalt, icy, and snowy road surface. Then, road point cloud is extracted using cloth simulation filtering and RANSAC algorithms, and abnormal noise points are removed based on Gaussian filtering. The road surface is divided into different regions according to the variation of point cloud reflectivity with distance and incident angle, and features are extracted accordingly. A road recognition model is constructed based on the deep neural network and trained by the collected dataset. Finally, the friction coefficient of the road ahead is determined based on the statistical experience of road material and peak friction coefficient. The test results show that the proposed algorithm achieves road type recognition accuracy of over 99.3%, with an average running cycle of 55ms, enabling real-time and high-precision estimation of the road peak friction coefficient.

Key words: road friction coefficient, LiDAR point cloud, cloth simulation filtering, RANSAC, deep neural network, Gaussian filtering, road type recognition