Abstract

In order to implement real‐time detection of passengers in subway stations, this paper proposes the SPDNet based on YOLOv4. Aiming at the low detection accuracy of passengers in the subway station due to uneven light conditions, we introduce the attention mechanism CBAM to recalibrate the extracted features and improve the robustness of the network. For the crowded areas in the subway station, we use the K‐means++ algorithm to generate anchors that are more consistent with the passenger aspect ratio based on the dataset KITTI, which mitigates the missing caused by the incorrect suppression of true positive boxes by the Nonmaximum Suppression algorithm. We train and test our SPDNet on the KITTI dataset and prove the superiority of our method. Then, we carry out transfer learning based on the subway surveillance video dataset collected by ourselves to make it conform to the distorted passenger targets under the angle of the surveillance camera. Finally, we apply our network in a Beijing subway station and achieve satisfactory results.

Highlights

  • As an important method to relieve the pressure of urban traffic at present, the intelligent construction of subway has attracted more and more attention in the industry

  • Considering that the YOLOv4 network already meets the real-time requirements of subway passenger detection tasks and has slightly higher accuracy and robustness compared to the current YOLOv5 network, the more stable model YOLOv4 is selected as the baseline for this network

  • This research is aimed at realizing a real-time detection method of subway station passengers based on surveillance systems, so as to better serve production and life

Read more

Summary

Introduction

As an important method to relieve the pressure of urban traffic at present, the intelligent construction of subway has attracted more and more attention in the industry. Realtime detection of passengers based on a large number of video surveillance equipment in subway stations plays an important role in the construction of “Smart Subway.”. For urban rail transit dispatching, real-time detection of passengers in the station can be used to grasp the density of passenger flow in the station in time, which can support the prediction of passenger flow [1]. For the security in subway stations, accurate and rapid detection of passengers can effectively monitor the gathering situation of the crowds and facilitate security dispatch to prevent possible safety accidents such as stampede. The passenger detection method for the scene in the subway station is still relatively blank, and we urgently need a passenger detection method to solve such problems

Objectives
Methods
Findings
Conclusion
Full Text
Paper version not known

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call