Songdo Traffic: High-Accuracy Vehicle Trajectories from Drone Footage

Songdo Traffic is a large-scale dataset of approximately 700,000 georeferenced vehicle trajectories extracted from high-altitude bird's-eye view drone footage. The trajectories include WGS84 coordinates, kinematics, estimated vehicle dimensions, class, and lane assignment, sampled at 29.97 points per second. It was created by author rfonod and last updated on the platform in June 2026.

Use Cases

• Modeling traffic flow and congestion patterns based on high-frequency trajectory data.
• Training computer vision models for vehicle detection and classification based on drone footage.
• Developing algorithms for lane-keeping and road assignment in autonomous driving systems based on georeferenced paths.
• Analyzing vehicle kinematics such as speed and acceleration for safety and infrastructure planning.

Strengths

• Large scale with approximately 700,000 individual vehicle trajectories.
• High temporal resolution with data sampled at 29.97 points per second.
• Includes multiple data modalities per trajectory: WGS84 coordinates, kinematics, vehicle dimensions, and class.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The specific geographic and temporal coverage of the data collection is not detailed in the provided metadata.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

rfonod on Hugging Face

Collection Method

Extracted from high-altitude bird's-eye view drone footage.

Freshness

Last updated 2026-06-25 16:33:27; freshness should be verified.

Geography

Songdo, a smart city (inferred from title).