一种改进的乘客走行时间测量方法研究
The Design and Application Research of Subway Train Passenger Travel Time Measuring Method
李辉 1石旭 1李天宇 1姚世严 1郑剑飞1
作者信息
- 1. 北京轨道交通路网管理有限公司,北京 100101
- 折叠
摘要
乘客走行时间是乘客步行通过车站内特定路线所花费的时间,该时间在车站客流风险调查、清分模型验证调查中发挥重要作用.现有检测方法存在过采样或欠采样问题,过采样会降低准确率,欠采样会降低召回率.为了提高准确率和召回率,文章提出了一种基于行人跟踪的走行时间测量方法,该方法根据乘客轨迹进行单一采样,从而避免了过采样和欠采样,最终大幅提升测量方法的准确率和召回率.经充分的实验验证,在摄像头角度合适的情况下,该测量方法可以达到98%的准确率和52%的召回率,满足多项业务的要求.
Abstract
The time for passengers to walk is the time spent walking through a specific route in the station,which is used in the station passenger wind risk investigation,clearing model verification,and trial.Existing detection methods have over sampling or under sampling problems,oversampling reduces accuracy,and under sampling reduces recall.In order to improve the accuracy and recall rate,this paper proposes a walking time measurement method based on pedestrian tracking,which performs a single sampling according to the passenger trajectory,thereby avoiding over sampling and under sampling,and ultimately greatly improving the accuracy and recall of the measurement method.Sufficient experimental verification shows that the measurement method can achieve 98%accuracy and 52%recall rate when the camera angle is suitable,meeting the requirements of multiple services.
关键词
视频分析/行人跟踪/行人重识别/走行时间/智慧地铁Key words
video analysis/travel time measuring/deep learning/smart metro引用本文复制引用
基金项目
北京市基础设施投资有限公司科研项目(2021-26-02-13)
出版年
2024
NETL
NSTL
万方数据