汽车工程 ›› 2023, Vol. 45 ›› Issue (6): 1031-1039.doi: 10.19562/j.chinasae.qcgc.2023.06.013

所属专题: 底盘&动力学&整车性能专题2023年

• • 上一篇    下一篇

基于轮胎特征点的并行大型车辆朝向角计算

赵嘉豪1,齐志权1(),齐智峰2,王皓2,何磊2   

  1. 1.北京理工大学机械与车辆学院,北京  100081
    2.毫末智行科技有限公司,北京  100192
  • 收稿日期:2022-11-29 修回日期:2023-01-02 出版日期:2023-06-25 发布日期:2023-06-16
  • 通讯作者: 齐志权 E-mail:qizhiquan@bit.edu.cn
  • 基金资助:
    国家自然科学基金(52002025)

Calculation of Heading Angle of Parallel Large Vehicle Based on Tire Feature Points

Jiahao Zhao1,Zhiquan Qi1(),Zhifeng Qi2,Hao Wang2,Lei He2   

  1. 1.School of Mechanical and Vehicle Engineering,Beijing Institute of Technology,Beijing  100081
    2.Haomo Technology Co. ,Ltd. ,Beijing  100192
  • Received:2022-11-29 Revised:2023-01-02 Online:2023-06-25 Published:2023-06-16
  • Contact: Zhiquan Qi E-mail:qizhiquan@bit.edu.cn

摘要:

针对自动驾驶环境感知中并行大型车辆朝向角预测结果稳定性差的问题,本文中提出一种新的方法进行并行大型车辆朝向角计算。首先,基于相机逆投影模型提出根据轮胎特征点图像位置计算目标车辆朝向角的计算方法。然后,在现有单目3D目标检测模型上增加并训练子分支用于进行大型车辆轮胎特征点的检测。最后,对本文算法进行可视化验证。结果表明,该方法可得到准确的并行大型车辆朝向角,具有比单目3D目标检测模型更好的稳定性。

关键词: 自动驾驶, 环境感知, 朝向角计算, 深度学习, 目标检测

Abstract:

For the poor stability of the prediction results of heading angle of parallel large vehicles in autonomous driving environment perception, this paper proposes a new method for the calculation of heading angle of parallel large vehicle. Firstly, a method based on camera inverse projection model to calculate the heading angle of target vehicle according to the location of feature points of tires in the image is proposed. Subsequently, a sub-branch is added and trained on the existing monocular 3D object detection model for the detection of feature point of tire of large vehicles. Finally, the algorithm in this paper is visually verified. The results show that the method can obtain accurate heading angle of parallel large-scale vehicle, and has better stability than the monocular 3D object detection model.

Key words: autonomous driving, environment perception, heading angle calculation, deep learning, object detection