A Pilot Experimental Project for Predicting Pedestrian Flows Using Computer Vision and Deep Learning

Details

• Creators:
  Han, Chaeyeon ; Chivilkar, Shreya ; Guhathakurta, Subhrajit
• Corporate Creators:
  Center for Understanding Future of Travel Behavior and Demand (TBD) National University Transportation Center (UTC)
• Corporate Contributors:
  United States. Department of Transportation. University Transportation Centers (UTC) Program ; United States. Department of Transportation. Office of the Assistant Secretary for Research and Technology
• Subject/TRT Terms:
  Neural Networks Pedestrian Counts Pedestrian Flow Pedestrian Traffic Route Choice
• Resource Type:
  Tech Report
• Geographical Coverage:
  United States
• Edition:
  Final Report, 2023-2024
• Corporate Publisher:
  Center for Understanding Future of Travel Behavior and Demand (TBD) National University Transportation Center (UTC)
• Abstract:
  This study presents a pilot project for predicting pedestrian flows using video-based computer vision and deep learning techniques. Two custom datasets were collected on the Georgia Tech campus, featuring video recordings of multiple pedestrian pathways along with annotated pedestrian counts and directional flow information. A methodological framework was developed incorporating Convolutional Neural Networks (CNNs) for spatial pattern recognition, Graph Convolutional Networks (GCNs) to model spatial relationships among distributed sensors, and a temporal lag optimization procedure to account for delayed influence from distant recorders. Experimental results demonstrate that CNNs effectively estimate and predict short-term pedestrian flow with high accuracy in smaller, localized environments (ASPED v.a), particularly when pedestrian count data is included. In larger and more complex networks (ASPED v.c), GCNs yield improved predictive performance but also exhibit challenges such as over-smoothing. The study concludes with a discussion on the potential for future multimodal sensing deployments and outlines limitations related to generalizability and scalability of the proposed approach.
• Format:
  PDF
• Funding:
  69A3552344815, 69A3552348320
• Collection(s):
  University Transportation Centers
• Main Document Checksum:
  urn:sha-512:572a98f9e8ad124516f3dc3c57c1379448addd85aa6e906039695bbc1ee7e80d47842645d9e072aa5c465648892039d26ae18e90ecc4273e02cbfb84b6114cd3
• Download URL:
  https://rosap.ntl.bts.gov/view/dot/87067/dot_87067_DS1.pdf
• File Type:
  [PDF - 2.55 MB ]