• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2025, Vol. 61 ›› Issue (13): 80-95.doi: 10.3901/JME.2025.13.080

• 特邀专栏:价值链协同赋能的复杂制造系统:趋势、技术与挑战 • 上一篇    

扫码分享

面向隧道场景的智能车雷达与视觉协同感知研究

孙念怡1, 赵津1, 黄磊1, 王广玮1,2   

  1. 1. 贵州大学机械工程学院 贵阳 550025;
    2. 清华大学车辆与运载学院 北京 100084
  • 收稿日期:2024-06-30 修回日期:2025-01-30 发布日期:2025-08-09
  • 作者简介:孙念怡,女,博士研究生。研究方向为空地协同下的动态目标跟踪。E-mail:gs.nysun21@gzu.edu.cn;赵津(通信作者),男,博士,教授,博士研究生导师。研究领域为智能网联汽车。E-mail:zhaoj@gzu.edu.cn;黄磊,男,硕士研究生。研究方向为自动驾驶环境感知。E-mail:huang1998lei@163.com;王广玮,男,副教授。研究方向为自动驾驶与智能控制。E-mail:gwwang@gzu.edu.cn
  • 基金资助:
    国家自然科学基金(51965008)、黔科合平台人才CXTD[2022]009、黔科合平台人才(GCC[2023]016)资助项目。

Tunnel Scene Oriented Intelligent Vehicle Radar Vision Cooperative Sensing Research

SUN Nianyi1, ZHAO Jin1, HUANG Lei1, WANG Guangwei1,2   

  1. 1. School of Mechanical Engineering, Guizhou University, Guiyang 550025;
    2. School of Vehicle and Mobility, Tsinghua University, Beijing 100084
  • Received:2024-06-30 Revised:2025-01-30 Published:2025-08-09

摘要: 相机与毫米波雷达被广泛应用于智能车感知系统,但在隧道中,由于环境的影响对相机和毫米波雷达的感知带来了挑战。隧道出入口的光线变化快、光照条件差,使得相机在观测目标特性时出现较大波动,导致目标检测失败。而隧道属于半封闭场景、噪声反射强,使得毫米波雷达难以区分直接反射回来的信号和经过多次反射后回来的信号,导致将虚假目标的信号识别为真实目标。因此,面向隧道场景,对出入口处因光照突变导致相机图像质量不佳、细节丢失等问题展开研究,提出了自适应曝光控制模型来调整相机的曝光时间。该模型通过分析图像帧中不同语义类别的特征点数量随曝光时间变化的关系,以确保相机在快速变化的光照条件下仍能清晰成像。并创建了隧道感知数据集,有效提高模型感知能力的同时,弥补了隧道类数据集的稀缺。此外,针对车载毫米波雷达在隧道场景中面临多径回波干扰所导致的虚假目标问题,通过构建多径传播理论模型,分析雷达回波中潜在虚假目标位置以及能量衰减的特点,提出多径假目标消除策略对虚假干扰目标进行消除。最后,在相机与毫米波雷达融合关联中引入运动目标的角点光流估计,以提高相机与毫米波雷达协同感知的可靠性,并搭建实车平台在隧道场景中进行试验。结果表明,提出的协同感知算法相比于其他模型在检测准确率上提高了4.8%,有效抑制了隧道出入口光照突变条件下对智能车的感知影响,并减少毫米波雷达在隧道环境中多径回波的干扰,实现传感器之间信息的互补和优势互补,为智能车在隧道环境中的安全行驶提供了重要保障。

关键词: 隧道智能驾驶, 协同感知, 自适应曝光, 多径效应, 光照突变

Abstract: Tunnel scenes are characterized by rapid light changes, poor lighting conditions, and noise interference, etc. When the intelligent vehicle senses the tunnel environment, it is prone to omission and error detection, leading to traffic accidents. Therefore, for tunnel scenes, a cooperative perception system and dataset based on the fusion of camera and millimeter-wave radar were constructed, carries out research on the problems of poor camera image quality and loss of details due to sudden changes in illumination at tunnel entrances and exits, and proposes an adaptive exposure control model to adjust the exposure time of the camera. The model analyzes the relationship between the number of feature points of different semantic categories in an image frame as a function of exposure time to ensure that the camera can still image clearly under rapidly changing lighting conditions. In addition, for the vehicle-mounted millimeter-wave radar facing the false target problem caused by multipath echo interference in tunnel scenarios, the multipath propagation theory model is built to analyze the characteristics of potential false targets position and energy attenuation in the radar echo, and the multipath false-target elimination strategy is adopted to eliminate the false interference targets. Finally, the corner-point optical flow estimation of moving targets is introduced in the fusion correlation of camera and millimeter-wave radar to improve the reliability of camera and millimeter-wave radar co-sensing, and a real-vehicle platform is constructed to conduct experiments in a tunnel scenario. The results show that the detection accuracy of the proposed cooperative perception algorithm is increased by 4.8% compared with other models, and it has a better vehicle perception performance in tunnel scenarios, which provides an important guarantee for the safe driving of intelligent vehicles in tunnel environments.

Key words: tunnel autonomous driving, collaborative perception, adaptive exposure, multipath interference, sudden light changes

中图分类号: