汽车工程 ›› 2021, Vol. 43 ›› Issue (7): 971-977.doi: 10.19562/j.chinasae.qcgc.2021.07.003

• • 上一篇    下一篇

基于Residual BiLSTM网络的车辆切入意图预测研究

郭景华1,2(),肖宝平1,王靖瑶1,罗禹贡2,陈涛3,李克强2   

  1. 1.厦门大学航空航天学院,厦门 361005
    2.清华大学,汽车安全与节能国家重点实验室,北京 100084
    3.中国汽车工程研究院股份有限公司,重庆 401122
  • 收稿日期:2021-01-22 出版日期:2021-07-25 发布日期:2021-07-20
  • 通讯作者: 郭景华 E-mail:guojh@xmu.edu.cn
  • 基金资助:
    国家重点研发计划(2016YFB0100900);汽车安全与节能国家重点实验室开放课题(KF2011);中央高校基本科研业务费专项资金(20720190015)

Study on Vehicle Cut⁃in Intention Prediction Based on Residual BiLSTM Network

Jinghua Guo1,2(),Baoping Xiao1,Jingyao Wang1,Yugong Luo2,Tao Chen3,Keqiang Li2   

  1. 1.School of Aerospace Engineering,Xiamen University,Xiamen 361005
    2.Tsinghua University,State Key Laboratory of Automotive Safety and Energy,Beijing 100084
    3.China Automotive Engineering Research Institute Co. ,Ltd. ,Chongqing 401122
  • Received:2021-01-22 Online:2021-07-25 Published:2021-07-20
  • Contact: Jinghua Guo E-mail:guojh@xmu.edu.cn

摘要:

本文中根据中国实际道路特征,提出一种基于Residual BiLSTM网络的车辆切入意图预测模型,从切入车辆的轨迹信息和与自车的交互信息中提取切入特征,并采用softmax函数计算切入意图,分别为左车道保持、左车道插入、右车道插入和右车道保持的概率,最后利用中国复杂路况的自然驾驶数据集对预测模型进行训练和测试。结果表明,所提出的Residual BiLSTM车辆切入意图预测模型有明显优势,其准确率比LSTM提升8.2个百分点,且能较早地预测出车辆的切入意图,对提高自动驾驶车辆的决策规划能力和安全性具有重要的意义。

关键词: 自动驾驶, 切入意图预测, 深度学习, 交互信息, 自然驾驶数据

Abstract:

A vehicle cut?in intention prediction model based on Residual BiLSTM network is proposed in this paper according to the real road features in China. The cut?in features are extracted from the trajectory information of the cut?in vehicle and its interaction information with ego vehicle, and the softmax function is used to calculate the cut?in intention, e.i. the probability of left lane keeping, left lane cut?in, right lane cut?in or right lane keeping respectively. Finally, the prediction model is trained and tested with the naturalistic driving data set on the complex roads in China. The results show that the Residual BiLSTM model proposed has obvious advantages in cut?in intention prediction, with an accuracy 8.2 percentage points higher than that of LSTM model, and can predict the vehicle cut?in intention earlier, being of great significance in enhancing the decision?making and planning ability and safety of autonomous vehicles.

Key words: autonomous driving, cut?in intention prediction, deep learning, interaction information, naturalistic driving data