长沙理工大学学报(自然科学版)
基于航拍轨迹的新能源汽车危险驾驶行为研究
CSTR:
作者:
作者单位:

(1. 湖南大学 土木工程学院 ,湖南 长沙 410082;2. 湖南大学 综合交通研究中心 ,湖南 长沙 410082;3. 智慧道路与绿色交通湖南省普通高等学校重点实验室 ,湖南 长沙 410082;4.内蒙古电力勘测设计院有限责任公司 ,内蒙古 呼和浩特 010020)

作者简介:

通讯作者:

陈升(2001—)(ORCID:0009-0001-2744-108X),男,主要从事新能源汽车安全与运营、城市轨道交通运营组织方面的研究。E-mail:cs@hnu.edu.cn

中图分类号:

U461.91;TP391.4

基金项目:

中央高校基本科研业务费专项资金项目(531118010636)第一作者:刘晨辉(1987—)(ORCID:0000-0003-2993-9022),男,教授,主要从事新能源汽车安全与运营、公共交通(城市轨道交通)组织、自然驾驶行为、智能交通控制与仿真方面的研究。E-mail:chenhuiliu@hnu.edu.cn


Research on high -risk driving behavior of new energy vehicles based on aerial trajectories
Author:
Affiliation:

(1. College of Civil Engineering , Hunan University , Changsha 410082, China;2. Comprehensive Transportation Research Center , Hunan University , Changsha 410082, China;3. Hunan Provincial University Key Laboratory of Smart Road and Green Transportation , Changsha 410082, China; 4. Inner Mongolia Electric Power Survey & Design Institute Co ., Ltd., Hohhot 010020, China)

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    【目的】随着新能源汽车保有量的迅速攀升,其涉及的交通事故也显著增多。交通事故主要由人为因素导致。本研究旨在构建一种适用于新能源汽车的危险驾驶行为量化识别体系,以及时识别驾驶员的危险驾驶行为并预防事故。【方法】首先,对无人机高空拍摄的车辆轨迹视频进行提取与分析,构建了新能源汽车自然驾驶数据集。其次,确定超速、急变速与危险变道 3类常见危险驾驶行为的特征参数阈值,通过计算特征值、相关性分析,构建了新能源汽车危险驾驶行为谱。然后,基于该行为谱,结合文献与视频观察确定危险驾驶行为比例,并对数据进行混合采样均衡处理。最后,通过对比随机森林、KNN、支持向量机、XGBoost 与AdaBoost 等模型的识别精度,筛选最优识别模型。【结果】基于混合采样均衡处理的 AdaBoost 算法被确定为最优识别模型,其精确率达 97.6%,召回率达 100.0%,AdaBoost 算法能有效识别潜在的危险驾驶行为。对构建的危险驾驶行为谱进行分析,结果显示超速、急变速、危险变道的权重分别为 0.401、0.268与0.331,其中超速行为对行车安全影响最大。【结论】本研究构建的识别体系能有效识别新能源汽车的危险驾驶行为,为交通安全管理提供数据支持与理论基础。

    Abstract:

    [Purposes ] With the rapid increase in the number of new energy vehicles,the traffic accidents involving them also significantly increase.Traffic accidents are mainly caused by human factors.This study aims to construct a quantitative recognition system for high -risk driving behaviors applicable to new energy vehicles,to promptly identify drivers ’ high -risk driving behaviors and prevent accidents.[Methods] Firstly,unmanned aerial vehicles were utilized to capture vehicle trajectory videos from high altitudes,and after the extraction and analysis of the videos,a naturalistic driving dataset of new energy vehicles was constructed.Secondly,the characteristic parameter thresholds of three common high -risk driving behaviors,including speeding,abrupt speed change,and dangerous lane -changing,were determined.A high -risk driving behavior spectrum of new energy vehicles was constructed by calculating the characteristic values and conducting a correlation analysis.Then,based on this behavior spectrum,the proportion of high -risk driving behaviors was determined combined with literature and video observation,and the data were balanced by hybrid sampling.Finally,the optimal recognition model was selected by comparing the recognition accuracies of models such as random forest,KNN,support vector machine,XGBoost,and AdaBoost.[Results] The AdaBoost algorithm based on hybrid sampling is determined as the optimal recognition model,with a precision of 97.6% and a recall of 100.0%,and the AdaBoost algorithm can effectively identify potential high -risk driving behaviors.The analysis of the constructed high -risk driving behavior spectrum shows that the weights of speeding,abrupt speed change,and dangerous lane -changing are 0.401,0.268,and 0.331,respectively,among which speeding behavior has the greatest impact on driving safety.[Conclusions ] The recognition system constructed in this study can effectively identify the high-risk driving behaviors of new energy vehicles,providing data support and a theoretical foundation for traffic safety management.

    参考文献
    相似文献
    引证文献
引用本文

刘晨辉,侯学睿,龙荣显,等.基于航拍轨迹的新能源汽车危险驾驶行为研究[J].长沙理工大学学报(自然科学版),2026,23(3):99-111.
LIU Chenhui, HOU Xuerui, LONG Rongxian, et al. Research on high -risk driving behavior of new energy vehicles based on aerial trajectories[J]. Journal of Changsha University of Science & Technology (Natural Science),2026,23(3):99-111.

复制
分享
相关视频

文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2026-02-09
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期: 2026-09-24
  • 出版日期:
文章二维码