交通运输系统工程与信息

交通运输系统工程与信息 ›› 2020, Vol. 20 ›› Issue (1): 40-46.

• 智能交通系统与信息技术 • 上一篇    下一篇

高速场景相邻前车驾驶行为识别及意图预测

张海伦，付锐*   

  1. 长安大学汽车学院，西安 710064
  • 收稿日期:2019-08-29 修回日期:2019-11-07 出版日期:2020-02-25 发布日期:2020-03-02
  • 作者简介:张海伦(1992-)，男，安徽马鞍山人，博士生.
  • 基金资助:

    国家重点研发计划/ National Key Research and Development Program of China(2018YFB1600501)；国家自然科学基金/National Natural Science Foundation of China(51775053)；长江学者与创新团队项目/Changjiang Scholars and Innovative Research Team in University(IRT_17R95).

Driving Behavior Recognition and Intention Prediction of Adjacent Preceding Vehicle in Highway Scene

ZHANG Hai-lun, FU Rui   

  1. School of Automobile, Chang'an University, Xi'an 710064, China
  • Received:2019-08-29 Revised:2019-11-07 Online:2020-02-25 Published:2020-03-02

PDF

摘要：

相邻前车的驾驶行为会影响后车，因此先进的辅助驾驶系统需具备识别前车驾驶行为的能力. 对高速场景下相邻前车换道行为进行研究，分别提出双层连续隐马尔可夫模型—贝叶斯生成分类器(CHMM-BGC)，以及基于双向长短时记忆网络(Bi-LSTM)的行为识别模型和意图预测模型. 采用自然驾驶数据集对模型的有效性进行测试验证. 实验分析表明：基于Bi-LSTM的行为识别模型相较于双层CHMM-BGC在平均识别率上提升了11.24%，两种行为识别模型均可在相邻前车换道过程的早期阶段识别换道行为；考虑相邻前车与周围环境车辆的交互作用，可使模型具有预测性，两种意图预测模型均可在车辆换道时刻前预测到驾驶人换道意图. 模型仿真计算时间可满足系统的实时性需求，为本车驾驶人预留出反应时间，为预测周围车辆行驶轨迹研究提供支持.

关键词: 交通工程, 行为识别, 意图预测, 连续隐马尔可夫模型, 双向长短时记忆网络

Abstract:

The driving behavior of the adjacent preceding vehicle in highway scene will have an impact on the vehicles behind. Advanced driver assistance systems (ADAS) should be equipped with the ability to recognize the lane changing behavior of the adjacent preceding vehicle. In this paper, the lane change behavior of adjacent preceding vehicle in highway scene was studied, the behavior recognition model and intention prediction model based on the two-layer continuous Hidden Markov Model-Bayesian generation classifier (CHMM-BGC) and the Bi- directional long short time memory network (Bi-LSTM) were proposed, respectively. The validity of the models was tested and verified using a natural driving data set. Experimental analysis shows that the average recognition rate of the behavior recognition model based on Bi-LSTM is 11.24% higher than that of two layers CHMM-BGC. Both behavioral recognition models can recognize lane change behavior of the adjacent preceding vehicle in the early stages of lane change process. Considering the interaction between the adjacent preceding vehicle and surrounding vehicles, the model can be predictive. Both intention prediction models can predict the driver's lane change intention before the lane change moment of the vehicle. The model simulation time satisfies the real-time requirement of the system. It can reserve the reaction time for the driver and provide support for the prediction of the surrounding vehicle trajectory research.

Key words: traffic engineering, behavior recognition, intention prediction, CHMM, Bi-LSTM

中图分类号: