City-scale Vehicle Trajectory Data from Traffic Camera Videos

Area of Science:

• Intelligent Transportation Systems
• Computer Vision
• Data Science

Background:

• Traditional vehicle trajectory datasets are limited by GPS bias and short, incomplete paths.
• Widespread traffic camera deployment offers an opportunity for comprehensive data collection.
• Existing datasets hinder reliable applications in traffic surveillance, prediction, and control.

Purpose of the Study:

• To develop and release the first comprehensive, city-scale vehicle trajectory dataset.
• To reconstruct vehicle paths across multiple traffic cameras.
• To provide a reliable data resource for intelligent transportation systems research.

Main Methods:

• A trajectory recovery framework was applied to city-scale traffic camera video data.
• Vehicles were identified across cameras, and their paths were reconstructed.
• The dataset was validated through method evaluation, case visualization, and comparison with external statistics.

Main Results:

• Approximately 5 million vehicle trajectories were recovered from over 3000 traffic cameras in two metropolises.
• The recovered trajectories demonstrated consistency and reliability.
• The dataset provides near-complete coverage of city vehicle movements.

Conclusions:

• The released dataset is a valuable resource for advancing intelligent transportation systems.
• It enables new research in traffic dynamics, network resilience, and planning.
• This work overcomes limitations of traditional trajectory data collection.