汽车工程

汽车工程 ›› 2020, Vol. 42 ›› Issue (1): 52-58.doi: 10.19562/j.chinasae.qcgc.2020.01.008

• • 上一篇    下一篇

驾驶员避撞转向行为的改进K-means聚类与识别*

赵治国, 冯建翔, 周良杰, 王凯, 胡昊锐, 张海山, 宁忠麟   

  1. 1.同济大学新能源汽车工程中心,上海 201804;
    2.同济大学汽车学院,上海 201804
  • 收稿日期:2019-02-21 发布日期:2020-01-21
  • 通讯作者: 冯建翔,硕士研究生,E-mail:1610850@tongji.edu.cn
  • 基金资助:
    *国家自然科学基金联合基金项目(U1564208)资助

K-means++ Clustering and Recognition of Driver'sCollision Avoidance Steering Behavior

Zhao Zhiguo, Feng Jianxiang, Zhou Liangjie, Wang Kai, Hu Haorui, Zhang Haishan, Ning Zhonglin   

  1. 1.Clean Energy Automotive Engineering Center, Tongji University, Shanghai 201804;
    2.School of Automotive Studies, Tongji University, Shanghai 201804
  • Received:2019-02-21 Published:2020-01-21

PDF (PC)

522

赞

摘要： 本文中根据不同工况驾驶员转向行为数据,提出了基于驾驶员避撞转向行为特征的聚类算法。首先搭建驾驶模拟器,采集了定半径转向、常规换道和紧急避撞转向工况下的驾驶行为数据,通过对比正常行驶和紧急避障工况下驾驶员转向行为数据,定性分析了紧急避撞转向特点。之后,利用皮尔逊相关系数法分析了描述驾驶员转向行为的观测变量与紧急避撞转向行为的相关性,得出转向盘转速与转向工况的相关性最高。接着,以转向盘转速作为聚类特征参数,利用改进K均值(K-means++)聚类方法对转向行为数据进行了聚类,将转向行为划分为正常转向和紧急避撞转向,实现了紧急避撞转向工况的识别。最后,通过实车试验验证了所提出的紧急避撞转向行为K-means++聚类方法可有效识别驾驶员紧急避撞转向行为,聚类精度达96.7%。

关键词: 避撞转向行为, 相关性分析, 改进K均值聚类, 识别

Abstract: A clustering algorithm based on driver's collision avoidance steering behavior's characteristics is proposed according to driver's steering behavior data under different working conditions in this paper. Firstly a driving simulator is built, on which the driving behavior data are collected under the conditions of fixed radius steering, conventional lane change and emergency collision avoidance steering. The features of emergency collision avoidance steering are qualitatively analyzed by comparing the steering behavior data of normal driving with that in emergency collision avoidance conditions. Then the Pearson correlation coefficient method is used to analyze the correlation between the measurement variables of driver's steering behavior and emergency collision avoidance steering behavior, with a result showing that the steering condition is most correlated to the rotational speed of steering wheel. After that, with the rotational speed of steering wheel as the clustering characteristic parameter, clustering is conducted on steering behavior data by using K-means++ algorithm, and the steering behaviors are divided into normal steering (including fixed-radius steering and lane change steering) and emergency collision avoidance steering, achieving the recognition of emergency collision avoidance steering. Finally real vehicle verification test is performed and the results indicate that the K-means++ clustering algorithm proposed can effectively identify the steering behavior of driver for emergency collision avoidance with a clustering accuracy up to 96.7%

Key words: collision avoidance steering behavior, correlation analysis, K-means++ clustering, recognition