汽车工程 ›› 2022, Vol. 44 ›› Issue (2): 153-160.doi: 10.19562/j.chinasae.qcgc.2022.02.001

所属专题: 智能网联汽车技术专题-规划&控制2022年

• •    下一篇

人机混驾环境下基于深度学习的车辆切入轨迹预测

郭景华1(),何智飞1,罗禹贡2,李克强2   

  1. 1.厦门大学航空航天学院,厦门  361005
    2.清华大学车辆与运载学院,北京  100084
  • 收稿日期:2021-09-06 修回日期:2021-11-06 出版日期:2022-02-25 发布日期:2022-02-24
  • 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);国家自然科学基金(61803319);厦门市重大科技项目(3502Z20201015)

Vehicle Cut-in Trajectory Prediction Based on Deep Learning in a Human-machine Mixed Driving Environment

Jinghua Guo1(),Zhifei He1,Yugong Luo2,Keqiang Li2   

  1. 1.School of Aerospace Engineering,Xiamen University,Xiamen  361005
    2.School of Vehicle and Mobility,Tsinghua University,Beijing  100084
  • Received:2021-09-06 Revised:2021-11-06 Online:2022-02-25 Published:2022-02-24
  • Contact: Jinghua Guo E-mail:guojh@xmu.edu.cn

摘要:

为保证人机混驾环境下自动驾驶汽车的安全行驶,对周围车辆切入轨迹的预测至关重要。本文首先使用Savitzky-Golay滤波器对大规模采集的自然驾驶数据进行去噪处理,根据准则提取车辆切入片段,建立符合中国道路状况的车辆切入数据集。其次,发挥Bi-LSTM网络能充分利用上下文信息的优点和in-out快捷连接有效减少梯度消失和网络退化的能力,提出了一种基于深度学习的改进型Bi-LSTM神经网络,来预测车辆的切入轨迹,在双向长短时记忆网络的基础上引入快捷连接,并综合考虑了自车对周边车辆切入的影响。利用自然驾驶数据集和NGSIM数据集进行试验验证,结果表明,本文提出的改进型Bi-LSTM预测模型的轨迹预测效果明显优于其他方法,对增强自动驾驶汽车的安全性具有重要的意义。

关键词: 自动驾驶, 切入轨迹预测, 人机混驾, 深度学习, 改进型Bi-LSTM

Abstract:

In order to ensure the safe driving of autonomous vehicles in a human-machine mixed driving environment, the prediction of the cut-in trajectory of surrounding vehicles is of paramount importance. Firstly in this paper, the Savitzky-Golay filter is used for the denoise treatment of large-scale natural driving data collected, the vehicle cut-in fragments are extracted based on criteria, and the vehicle cut-in dataset consistent with the road conditions in China is established. Secondly, by giving play to the advantage of Bi-LSTM in fully utilizing context and the ability of in-out swift connection in effectively reducing gradient disappearance and network degeneration, an improved bi-directional long short-term memory (Bi-LSTM) neural network based on deep learning is proposed to predict the trajectory of cut-in vehicle, and the swift connection is introduced on the basis of Bi-LSTM with comprehensive consideration of the effects of ego vehicle on the cut-in of surrounding vehicles. A test verification is conducted on the natural driving dataset and NGSIM dataset with a result showing that the trajectory prediction effects of the improved Bi-LSTM prediction model proposed are significantly better than other methods, having important significance in enhancing the safety of autonomous vehicles.

Key words: autonomous driving, cut-in trajectory prediction, human-machine mixed driving, deep learning, improved Bi-LSTM