欢迎访问《汽车安全与节能学报》，
汽车安全与节能学报

汽车安全与节能学报 ›› 2024, Vol. 15 ›› Issue (4): 602-609.DOI: 10.3969/j.issn.1674-8484.2024.04.017

• 智能驾驶与智慧交通 • 上一篇    

全触屏人机交互模式下驾驶人视觉分心时长预测

王畅(), 牛津(), 王一飞, 张雅丽*(), 马万良   

  1. 长安大学 汽车学院，西安 710064，中国
  • 收稿日期:2024-01-24 修回日期:2024-03-20 出版日期:2024-08-31 发布日期:2024-09-10
  • 通讯作者: *张雅丽，讲师。E-mail：zhangyali@chd.edu.cn。
  • 作者简介:王畅（1984—），男（汉），湖南，教授。E-mail：wangchang@chd.edu.cn。
    牛津（2001—），男（汉），河南，硕士研究生。E-mail：niujin@chd.edu.cn。
  • 基金资助:
    陕西省自然科学基础研究计划资助项目(2023-JC-YB-596);陕西省重点研发计划资助项目(2023-YBGY-035)

Prediction of driver visual distraction duration under a human-machine interaction mode with full-touch-screen

WANG Chang(), NIU Jin(), WANG Yifei, ZHANG Yali*(), MA Wanliang   

  1. Author Institute School of Automobile, Chang’an University, Xi’an 710064, China
  • Received:2024-01-24 Revised:2024-03-20 Online:2024-08-31 Published:2024-09-10

RichHTML

6

PDF

66

摘要：

为了准确预测全触屏人机交互模式下驾驶人视觉分心时长，建立了考虑驾驶人分心风格的视觉分心时长预测模型。采用实车道路试验所采集的多车速视觉分心数据，基于击键水平模型（KLM）建立一组适用于全触屏车载设备的基本操作单元，以及以各基本操作单元数量与车速作为输入特征的视觉分心时长预测随机森林模型（RF），并采用自组织映射（SOM）算法将驾驶人聚类为谨慎型、正常型、激进型3类分心风格表征驾驶人视觉分心特性差异，将该参数作为输入特征添加至原模型实现优化。结果表明：优化后模型具有最良好的预测精度，测试集的均方误差、平均绝对误差和决定系数R2分别为2.4148 s2、1.0371 s、0.9345，较原模型分别提升30.52%、11.8%、3.18%；模型性能显著优于线性回归模型与XGBoost模型。研究结果可用于协助车载辅助驾驶系统适时警示或实施干预以降低分心所致的追尾风险，并为交互界面设计提供指导意见。

关键词: 汽车主动安全, 视觉分心, 击键水平模型（KLM）, 随机森林模型（RF）, 分心风格

Abstract:

A visual distraction duration prediction model considering the driver's distraction style was established to accurately predict the driver's visual distraction duration in the full touch-screen human-computer interaction mode. Based on the multi-speed visual distraction data collected from the real car road test, a set of basic operation units suitable for full touch screen vehicle-mounted devices was constructed based on the Keystroke Level Model (KLM), and a Random Forest Model (RF) for visual distraction duration prediction with the number of each basic operation units and vehicle speed as input features was established. The Self-Organizing Map (SOM) Algorithm was used to cluster drivers into three types of distraction styles: Cautious, normal and aggressive to characterize the differences in drivers' visual distraction characteristics, and the parameters were added to the original model as input features to realize optimization. The results show that the optimized model has the best prediction accuracy. The mean square error, mean absolute error and determination coefficient R2 of the test set are 2.414 8 s2, 1.037 1 s and 0.934 5, respectively, which are 30.52%, 11.8% and 3.18% higher than those of the original model. The model performance is significantly better than the linear regression model and XGBoost model. The results can be used to assist in-vehicle driver assistance systems in timely warning or implementing intervention to reduce the risk of rear-end collision caused by distraction, and provide guidance for the design of interactive interfaces.

Key words: vehicle active safety, visual distraction duration, Keystroke Level Model (KLM), Random Forest(RF), distraction style

中图分类号: