Real-Time Enumeration of Metro Passenger Volume Using Anchor-Free Object Detection Network on Edge Devices
Open Access
- 1 January 2021
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 9, 21593-21603
Abstract
Enumeration of metro passenger volume is essential in providing effective passenger guidance and improving the usage rate of each carriage. However, existing methods cannot provide the accurate number of alighting and boarding passengers at each gate on the platform. For existing visual methods, the occlusion problem seriously affected the results caused by the angle of view. In this study, we introduce a real-time metro passenger volume enumerating algorithm that is simple, effective, and fast enough to run on edge devices mounted above the platform gate. First, we capture videos from the cameras and design an anchor-free object detection network called CircleDet to detect passengers’ heads. CircleDet predicts a circle to localize and bound the target instead of traditional bounding box. Then, we apply a simple but effective circle IoU-based method to identify and track passengers in the videos. CircleDet can achieve up to 111 frames per second (FPS) running on NVIDIA RTX 2080 and 7.8 FPS on an NVIDIA Jetson Nano device. The accuracy of enumeration is as high as 97.1% on our own metro object detection (MOD) dataset.Keywords
Funding Information
- China’s 13th Five-year Key Research and Development Plan for the Safety Guarantee Technology of Urban Rail Transit System (2016YFB1200402)
This publication has 29 references indexed in Scilit:
- CornerNet: Detecting Objects as Paired Keypoints2018
- YOLO9000: Better, Faster, Stronger2017
- Aggregated Residual Transformations for Deep Neural Networks2017
- SSD: Single Shot MultiBox Detector2016
- Faster R-CNN: Towards Real-Time Object Detection with Region Proposal NetworksIEEE Transactions on Pattern Analysis and Machine Intelligence, 2016
- You Only Look Once: Unified, Real-Time Object Detection2016
- Fast R-CNN2015
- Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation2014
- Histograms of Oriented Gradients for Human Detection2005
- Distinctive Image Features from Scale-Invariant KeypointsInternational Journal of Computer Vision, 2004