主办：陕西省汽车工程学会
ISSN 1671-7988  CN 61-1394/TH
创刊：1976年
汽车实用技术

汽车实用技术 ›› 2025, Vol. 50 ›› Issue (15): 24-29，47.DOI: 10.16638/j.cnki.1671-7988.2025.015.005

• 智能网联汽车 • 上一篇    

基于双目预瞄的路面三维信息获取

钱文斌 1,2，康绍鹏 1,2*，杨静 1,2，刘凯磊 1,2，强红宾 1,2   

  1. 1.江苏理工学院 机械工程学院； 2.江苏省先进流体动力与装备工程研究中心
  • 发布日期:2025-08-08
  • 通讯作者: 康绍鹏
  • 作者简介:钱文斌（1999－），男，硕士研究生，研究方向为超低重载主动悬挂与双目预瞄
  • 基金资助:
    国家自然科学基金项目（51805228）；江苏省高等学校基础科学（自然科学）研究项目（22KJB460021， 23KJA460006）；常州市领军型创新人才引进培育项目（CQ20210093，CQ20220089）；江苏理工学院研究生 实践创新计划项目（XSJCX23_54）

Obtaining in Road Surface Three-Dimensional Information Based on Binocular Preview

QIAN Wenbin1,2, KANG Shaopeng1,2* , YANG Jing1,2 , LIU Kailei1,2 , QIANG Hongbin1,2   

  1. 1.School of Mechanical Engineering, Jiangsu University of Technology; 2.Jiangsu Provincial Engineering Research Center for Advanced Fluid Power and Equipment
  • Published:2025-08-08
  • Contact: KANG Shaopeng

PDF

摘要： 立体匹配在三维感知和场景理解中至关重要，但传统双目识别方法在无纹理和纹理重 复场景中易产生较大的像素误差，影响三维模型的精度。文章提出基于深度感知层次模块的 改进 DH-GwcNet，通过深度感知层次模块解决路面纹理问题。在 KITTI 2015 和路面重建数据 集（RSRD）上的验证表明，该网络可以显著降低视差图像素误差，并在通过视差图构建三维 模型、提取三维信息时，其不同路面下路面不平度准确率均超过 90%。该研究不仅为实现汽 车悬挂装置的主动控制奠定基础，同时也在一定程度上提升了汽车行驶平顺性。

关键词: 双目预瞄；视差图；路面识别；注意力模块

Abstract: Stereo matching is crucial in three-dimensional perception and scene understanding, but traditional binocular recognition methods are prone to significant pixel errors in textureless and textured repetitive scenes, which can affect the accuracy of three-dimensional models. This paper proposes an improved DH-GwcNet based on the depth perception hierarchy module, which solves the problem of road surface texture through the depth perception hierarchy module. The validation on the KITTI 2015 and road surface reconstruction dataset (RSRD) datasets shows that the network significantly reduces pixel errors in disparity maps, and when constructing three-dimensional models and extracting three-dimensional information through disparity maps, its accuracy in road surface roughness under different road surfaces exceeds 90%. This study not only lays the foundation for achieving active control of automotive suspension installation, but also improves the smoothness of vehicle driving to a certain extent.

Key words: binocular preview; disparity map; road surface recognition; attention module