Pedestrian Origin-Destination Estimation Based on Multi-Camera Person Re-Identification.

Area of Science:

• Intelligent Transport Systems
• Computer Vision
• Urban Planning

Background:

• Pedestrian origin-destination (O-D) estimates are crucial for managing pedestrian facilities, including flow simulation and crowd control.
• Existing O-D data collection methods (surveys, mobile sensing, smart cards) are often costly, time-consuming, or incomplete, especially for facilities without clear entrances/exits.

Purpose of the Study:

• To develop and evaluate an image-based framework for accurate pedestrian O-D estimation.
• To address the limitations of current O-D data collection techniques for pedestrian facilities.
• To leverage CCTV camera networks and image processing for improved pedestrian flow analysis.

Main Methods:

• Proposed an image-based O-D estimation framework utilizing CCTV camera coverage.
• Developed a method to identify individuals across disjoint camera views to generate O-D trajectories.
• Compared and improved state-of-the-art deep neural networks (DNNs) for person re-identification under varying congestion levels.
• Generated an O-D matrix from trajectories and calculated resident times.

Main Results:

• Accurate generation of pedestrian O-D trajectories by identifying individuals across multiple camera views.
• Improved performance of deep neural networks for person re-identification in pedestrian environments.
• Creation of an O-D matrix and calculation of resident times, providing data for facility improvement.

Conclusions:

• The proposed image-based framework offers a viable solution for comprehensive pedestrian O-D estimation.
• This approach enhances pedestrian facility management through accurate flow simulation and operational insights.
• The study highlights the potential of the 'Internet of Cameras' for intelligent transport infrastructure management and suggests areas for future research.