汽车工程 ›› 2023, Vol. 45 ›› Issue (9): 1666-1676.doi: 10.19562/j.chinasae.qcgc.2023.09.015

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

基于毫米波雷达与视觉融合的碰撞预警

李勇滔1,孙晨旭1(),郑伟光1,2,许恩永3,李育方3,王善超3   

  1. 1.广西科技大学,重型车辆零部件先进设计制造教育部工程研究中心,柳州 545616
    2.吉林大学汽车工程学院,长春 130000
    3.东风柳州汽车有限公司,柳州 545616
  • 收稿日期:2023-03-23 修回日期:2023-04-25 出版日期:2023-09-25 发布日期:2023-09-23
  • 通讯作者: 孙晨旭 E-mail:sunchenxu6519@foxmail.com
  • 基金资助:
    国家自然科学基金(52065013);广西重点研发计划项目(桂科AB21220052)

Collision Warning Based on Fusion of Millimeter Wave Radar and Vision

Yongtao Li1,Chenxu Sun1(),Weiguang Zheng1,2,Enyong Xu3,Yufang Li3,Shanchao Wang3   

  1. 1.Guangxi University of Science and Technology,Engineering Research Center of Ministry of Education for Advanced Design and Manufacturing of Heavy Vehicle Components,Liuzhou 545616
    2.College of Automotive Engineering,Jilin University,Changchun 130000
    3.Dongfeng Liuzhou Motor Company,Liuzhou 545616
  • Received:2023-03-23 Revised:2023-04-25 Online:2023-09-25 Published:2023-09-23
  • Contact: Chenxu Sun E-mail:sunchenxu6519@foxmail.com

摘要:

针对现有的毫米波雷达与视觉融合的碰撞预警算法误警率与漏警率较高等问题,提出了一种基于毫米波雷达与视觉融合的碰撞预警方法。首先基于距离速度阈值与生命周期方法对毫米波雷达数据进行预处理,并提出基于遗忘因子的自适应拓展卡尔曼滤波算法对目标进行追踪;利用加入改进的CBAM双通道注意力机制YOLOv5算法提高视觉识别的准确率。然后运用交并比的思想实现毫米波雷达与视觉的决策级融合。最后基于最小安全距离模型提出前碰撞预警策略。通过不同场景下的实车验证结果表明:该算法比单传感器算法更加准确,具有更好的鲁棒性。

关键词: 毫米波雷达, 深度视觉, 车辆识别, 碰撞预警, 自适应卡尔曼滤波, 注意力机制

Abstract:

For the problems of high false alarm rate and missed alarm rate of existing collision warning algorithms of millimeter wave radar and vision fusion, a collision warning method based on millimeter wave radar and vision fusion is proposed in this paper. Firstly, the distance-velocity threshold and life cycle methods are used to pre-process the millimeter wave radar data, and the adaptive extended Kalman filter algorithm based on forgetting factor is proposed to track the target, adding the improved CBAM two-channel attention mechanism YOLOv5 algorithm to improve the accuracy of visual recognition. Then the idea of cross-comparison is applied to realize the decision-level fusion of millimeter wave radar and vision. Finally, a forward collision warning strategy is proposed based on the minimum safe distance model. The results of real-vehicle tests under different scenarios show that the algorithm is more accurate and has better robustness than the single-sensor algorithm.

Key words: millimeter?wave radar, depth vision, vehicle identification, collision warning, adaptive Kalman filtering, attention mechanism