汽车工程 ›› 2022, Vol. 44 ›› Issue (7): 997-1008.doi: 10.19562/j.chinasae.qcgc.2022.07.006

所属专题: 智能网联汽车技术专题-感知&HMI&测评2022年

• • 上一篇    下一篇

数据机理混合驱动的交通车意图识别方法

赵健1,宋东鉴1,朱冰1(),吴杭哲2,韩嘉懿1,刘宇翔1   

  1. 1.吉林大学,汽车仿真与控制国家重点实验室,长春  130022
    2.中国第一汽车集团有限公司智能网联开发院,长春  130011
  • 收稿日期:2021-11-15 修回日期:2022-02-20 出版日期:2022-07-25 发布日期:2022-07-20
  • 通讯作者: 朱冰 E-mail:zhubing@jlu.edu.cn
  • 基金资助:
    国家自然科学基金-智能汽车人机并行控制冲突机理与协同共驾关键技术研究(51775235)

Traffic Vehicles Intention Recognition Method Driven by Data and Mechanism Hybrid

Jian Zhao1,Dongjian Song1,Bing Zhu1(),Hangzhe Wu2,column:Han Jiayi1,Yuxiang Liu1   

  1. 1.Jilin University,State Key Laboratory of Automotive Simulation and Control,Changchun  130022
    2.Intelligent Connected Vehicle Development Institute of China FAW Group Co. ,Ltd. ,Changchun  130011
  • Received:2021-11-15 Revised:2022-02-20 Online:2022-07-25 Published:2022-07-20
  • Contact: Bing Zhu E-mail:zhubing@jlu.edu.cn

摘要:

交通车意图识别对提升智能汽车决策规划性能具有重要意义。本文从驾驶行为生成机理角度分析了驾驶人换道过程的各阶段,分别建立了基于马尔可夫决策过程的驾驶人意图预测模型、基于动态安全场的换道可行性分析模型和基于双向多长短期记忆网络(Bi-LSTM)的车辆行为识别模型。通过融合具有明确时序关系的上述模型,提出了一种数据机理混合驱动的交通车意图识别方法,并利用NGSIM数据集进行模型训练和验证。结果表明该方法在交通车到达换道点前1.8 s处的识别准确率即超过90%,在换道点处识别准确率可达97.88%,具有较高的识别准确率和较长的提前识别时间。

关键词: 智能汽车, 意图识别, 马尔可夫决策过程, 数据机理混合驱动

Abstract:

Traffic vehicle intention recognition is of great significance to improve the performance of intelligent vehicle decision-making and planning. This paper analyzes each stage of the driver’s lane changing process from the perspective of the driving behavior generation mechanism, and establishes the driver’s intention prediction model based on Markov decision process (MDP), the lane changing feasibility analysis model based on the dynamic safety field, and the vehicle behavior recognition model based on bi-directional long short-term memory(Bi-LSTM). By combining the above-mentioned models with a clear temporal relationship, a traffic vehicle intention recognition method driven by data and mechanism hybrid is proposed, and the NGSIM data set is used for model training and verification. The results show that the recognition accuracy of the proposed method is over 90% at 1.8 s before the traffic vehicle reaching the lane changing point, and the accuracy is 97.88% at the lane changing point, which proves high recognition accuracy and long advance recognition time.

Key words: intelligent vehicles, intention recognition, Markov decision process, data and mechanism hybrid driven