中山大学学报自然科学版 ›› 2011, Vol. 50 ›› Issue (4): 56-60.

• 研究论文 • 上一篇    下一篇

基于视频车辆轨迹模型的交通事件自动检测方法研究

赵有婷,李熙莹,罗东华

  

  1. (中山大学智能交通研究中心∥广东省智能交通系统重点实验室,广东 广州510275)
  • 收稿日期:2011-01-04 修回日期:1900-01-01 出版日期:2011-07-25 发布日期:2011-07-25

Study on the Methods of Automatic Incident Detection based on the Video Vehicle Trajectory Model

ZHAO Youting, LI Xiying, LUO Donghua   

  1. (Research Center of Intelligent Transportation System∥Guangdong Provincial Key Laboratory of Intelligent Transportation System, Sun Yatsen University, Guangzhou 510275,China)
  • Received:2011-01-04 Revised:1900-01-01 Online:2011-07-25 Published:2011-07-25

摘要: 研究了车辆违章逆行、停驻、掉头、倒退、变道五类具有潜在危险的交通(违章)事件,并且运用了基于视频的交通事件自动检测技术所涉及的目标提取、车辆跟踪和事件理解与描述3个步骤实现交通事件的检测。着重研究并分析了车辆跟踪得到的行驶轨迹点,将复杂的车辆轨迹分解为前行、反行、停滞、斜行四类轨迹元素,并且根据4类轨迹元素对车辆的行驶行为进行数学建模,最后通过模型制定合理的检测算法。实验表明,该算法可以有效地区分正常车辆与事件车辆,能够快速准确地检测上述5类交通事件。

关键词: 交通事件检测, 目标提取, 车辆跟踪, 车辆行驶行为模型

Abstract: Five dangerous traffic incidents including vehicle retrograding, parking, turning round, reversing, and changing lanes are studied.Object detection, vehicle tracking, incident apprehension and description are used to detect incident automatically based on video. The great emphasis is placed on analyses and researches of the vehicles tracking trajectory. Complicated vehicle trajectory is separated into four elements: ahead, backward, stop, and left or right; then vehicle behavior model is built to analyze vehicles motion based on them. Finally, reasonable algorithms to detect above five traffic incidents are proposed. Experiments show that the normal and the abnormal vehicles can be distinguished, and the five traffic incidents can be detected quickly and effectively by the algorithms.

Key words: traffic incident detection, object detection, vehicle tracking, vehicle behavior model

中图分类号: