交通运输系统工程与信息

交通运输系统工程与信息 ›› 2022, Vol. 22 ›› Issue (1): 49-56.DOI: 10.16097/j.cnki.1009-6744.2022.01.006

• 智能交通系统与信息技术 • 上一篇    下一篇

基于多入侵线的视频车速检测方法

田会娟* a, b，刘嘉伟a, b，翟佳豪a, b，邓琳琳a, b   

  1. 天津工业大学，a. 电子与信息工程学院；b. 天津市光电检测技术与系统重点实验室，天津 300387
  • 收稿日期:2021-08-19 修回日期:2021-09-16 接受日期:2021-09-23 出版日期:2022-02-25 发布日期:2022-02-23
  • 作者简介:田会娟(1979- )，女，河北新乐人，副教授。
  • 基金资助:
    国家自然科学基金；天津市科技计划项目

Video-based Vehicle Speed Measurement Method Using Multiple Intrusion Lines

TIAN Hui-juan* a, b, LIU Jia-weia, b, ZHAI Jia-haoa, b, DENG Lin-lina, b   

  1. a. School of Electronic and Information Engineering; b. Tianjin Key Laboratory of Optoelectronic Detection Technology and System, Tiangong University, Tianjin 300387, China
  • Received:2021-08-19 Revised:2021-09-16 Accepted:2021-09-23 Online:2022-02-25 Published:2022-02-23
  • Supported by:
    National Natural Science Foundation of China(61504095)；Tianjin Science and Technology Plan Program(18ZXCLGX00090)

PDF

摘要： 为提高视频中车速检测的精度，提出一种基于多入侵线的视频车速检测方法。首先在视频中布设已知相对距离的多条入侵线，其次检测车辆经过每条入侵线时的帧数，最后结合帧数、 摄像机的采样时间、入侵线间的距离生成关于车速的概率密度函数模型以计算车速。通过构建仿真环境验证模型性能，仿真结果表明：减小摄像机的采样时间、增加入侵线数量、增大入侵线之间的距离可以提高模型性能，并且在不同检测条件下使用多入侵法进行车速检测的误差率都更低。采用Deepsort+YOLOv5目标跟踪算法实现视频中车速的检测，同时，在视频车速检测综合数据集BrnoCompSpeed上与主流车速检测方法进行实验对比，实验结果表明，该方法测量结果的平均误差率为1.40%，与主流视频车速检测方法相比精度更高。

关键词: 智能交通, 多入侵线, 模式识别, 车速检测, 机器视觉

Abstract: To improve the accuracy of vehicle speed detection in videos, this paper proposes a video- based vehicle speed detection method based on multi-intrusion lines. In the method, the multiple intrusion lines with known relative distance in the video are established, and the frames is detected when the vehicle passes through each intrusion line. The vehicle speed is calculated by the probability density function model which is generated by combining the frames, the sampling time of the camera, and the distance among intrusion lines. The performance of the model is verified by building simulation environment. The results show that the performance of the model can be improved by reducing the sampling time of the camera and increasing the number of intrusion lines and the distance among the intrusion lines. The method can decrease the error rate of detecting vehicle speed under different detection conditions. The Deepsort+ YOLOv5 target tracking algorithm is used to realize the speed detection of the vehicle in the video. At the same time, the method is compared with the mainstream video-based vehicle detection methods on the BrnoCompSpeed comprehensive dataset. The results show that the average error rate obtained by the method is 1.40%, which is lower than the mainstream video-based vehicle speed detection methods.

Key words: intelligent transportation, multiple intrusion lines, pattern recognition, vehicle speed detection, machine vision

中图分类号: