欢迎访问《汽车安全与节能学报》,

汽车安全与节能学报 ›› 2022, Vol. 13 ›› Issue (1): 104-111.DOI: 10.3969/j.issn.1674-8484.2022.01.010

• 智能驾驶与智慧交通 • 上一篇    下一篇

基于深度学习LSTM的智能车辆避撞模型及验证

房亮1,2(), 关志伟1(), 王涛1, 龚进峰3, 杜峰1,4   

  1. 1.天津职业技术师范大学 汽车与交通学院,天津 300222,中国
    2.天津职业大学 汽车工程学院,天津 300410,中国
    3.中国汽车技术研究中心,天津 300300,中国
    4.天津市智能交通技术工程中心,天津 300222,中国
  • 收稿日期:2021-05-30 修回日期:2021-10-20 出版日期:2022-03-31 发布日期:2022-04-02
  • 通讯作者: 关志伟
  • 作者简介:*关志伟,教授。 E-mail: zhiwguan@163.com
    房亮(1988–),男(汉族),河北,讲师。 E-mail:fangliang@tute.edu.cn
  • 基金资助:
    国家重点研发计划项目(2017YFB0102500);天津市人工智能科技重大专项(17ZXRGGX00070);天津市科技计划项目(21YDTPJC00350);天津职业大学科学研究基金项目(20181109)

Collision avoidance model and its validation for intelligent vehicles based on deep learning LSTM

FANG Liang1,2(), GUAN Zhiwei1(), WANG Tao1, GONG Jinfeng3, DU Feng1,4   

  1. 1. School of Automotive and Transportation, Tianjin University of Technology and Education, Tianjin 300222, China
    2. School of Automotive Engineering, Tianjin Vocational University, Tianjin 300410, China
    3. Automotive Technology and Research Center, Tianjin 300300, China
    4. Tianjin Intelligent Transportation Technology Engineering Center, Tianjin 300222, China
  • Received:2021-05-30 Revised:2021-10-20 Online:2022-03-31 Published:2022-04-02
  • Contact: GUAN Zhiwei

摘要:

建立了一种基于深度学习长短期记忆(LSTM)网络的智能车辆避撞模型。搭载虚拟测试驾驶(VTD)仿真软件的模拟驾驶器;针对跟车行驶、前车紧急制动这一危险工况,开展模拟驾驶实验;以相对距离、相对速度、前车减速度、碰撞时间、横向距离作为输入参数;通过未经训练的样本数据对模型迁移性进行验证,并与传统反向传播(BP)神经网络进行对比。结果表明:本模型自车减速度和方向盘转角预测的决定系数R2分别高于传统BP模型0.17和0.21。因而,本文模型对大数据样本具有更优的拟合优度,能够表征驾驶员实际避撞行为,在驾驶辅助系统应用中具有推广价值。

关键词: 智能车辆, 避撞, 模拟驾驶器, 深度学习, 长短期记忆(LSTM), 驾驶辅助系统

Abstract:

A collision avoidance model was established for intelligent vehicles based on a deep learning Long Short-Term Memory (LSTM) network. Some driving simulation experiments were carried out at the conditions of following car dangerous driving or front car emergency braking, through a driving simulation platform with a simulation software of Virtual Test Driving (VTD). The relative distances, the relative speeds, the decelerations of the preceding car, the collision times, and the lateral distances were taken as the input parameters. Model mobility was verified by an untrained sample data, and compared with a traditional Back Propagation (BP) neural network. The results show that the R2 values of the model are higher than that of the traditional BP model by 0.17 and 0.21, respectively for the predicational decelerations and the predicational steering wheel angles. Therefore, this model has a fitting goodness for the big data samples and can represent the driver actual collision-avoidance behavior, with a value in promoting application of driver assistance system.

Key words: intelligent vehicles, collision avoidance, driving simulators, deep learning, long short-term memory (LSTM), driver assistance systems

中图分类号: