基于Seq2Seq自编码器模型的交通事故实时检测与评价
作者:
作者单位:

1.上海海事大学;2.华设设计集团股份有限公司;3.上海电科智能系统股份有限公司

作者简介:

通讯作者:

中图分类号:

C934

基金项目:

国家自然科学基金项目(面上项目,重点项目,重大项目)


Real-time traffic accident detection and evaluation based on Seq2Seq and AutoEncode model
Author:
Affiliation:

shanghai maritime university

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:

    路侧检测设备可以精准获取交通流量和速度等数据,交管部门借此可以显著提升对交通异常的感知程度。通过分析交通状态和交通流数据特征,本文建立了一套基于交通流序列数据的交通事故实时检测系统和预警流程。首先,在交通状态感知方面,本文建立的Seq2Seq自编码模型并引入了Attention机制,实现了重要交通状态特征的捕捉。其次,在交通状态异常判定方面,利用Seq2Seq自编码器重构输入的序列数据,对比原始数据可得到结构重构误差,然后再根据设定的阈值实现交通预警等级的判定和交通事故的实时检测。最后,本文采用上海市延安高架数据并通过混淆矩阵评价方法论证了该交通事故实时检测模型的可行性。

    Abstract:

    Roadside monitoring equipment can obtain traffic flow and speed data more accurately, so that traffic management departments can significantly improve the perception of traffic anomalies. By analyzing the characteristics of traffic state and traffic flow data, this paper establishes a set of real-time traffic accident detection system and early warning process based on traffic flow sequence data.First of all, in terms of traffic state perception, this paper introduces the Attention mechanism to capture important traffic state features based on the Seq2Seq model. Secondly, in the aspect of abnormal determination of traffic state, the Auto-Encoder is used to realize the reconstruction of input sequence data, and the structural reconstruction error is obtained by comparing the original data, and then the real-time detection of traffic accidents and the classification of accident warning levels are realized according to the set threshold.Finally, this research demonstrates the feasibility of the real-time traffic accident detection model by using the data of Yan ’an elevated highway (Shanghai) and using the confusion matrix evaluation method.

    参考文献
    相似文献
    引证文献
引用本文
分享
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:2020-12-14
  • 最后修改日期:2021-06-07
  • 录用日期:2021-06-17
  • 在线发布日期: 2021-07-01
  • 出版日期: