Songdo Traffic: High Accuracy Georeferenced Vehicle Trajectories from a Large-Scale Study in a Smart City

Overview

The Songdo Traffic dataset delivers precisely georeferenced vehicle trajectories captured through high-altitude bird's-eye view (BeV) drone footage over Songdo International Business District, South Korea. Comprising approximately 700,000 unique trajectories, this resource represents one of the most extensive aerial traffic datasets publicly available, distinguishing itself through exceptional temporal resolution that captures vehicle movements at 29.97 points per second, enabling unprecedented granularity for advanced urban mobility analysis.

⚠️ Important: If you use this dataset in your work, please cite the following reference [1]:

Robert Fonod, Haechan Cho, Hwasoo Yeo, Nikolas Geroliminis (2025). Advanced computer vision for extracting georeferenced vehicle trajectories from drone imagery, arXiv preprint arXiv:2411.02136.

(Note: This manuscript shall be replaced by the published version once available.)

Dataset Composition

The dataset consists of four primary components:

Trajectory Data: 80 ZIP archives containing high-resolution vehicle trajectories with georeferenced positions, speeds and acceleration profiles, and other metadata.

Orthophoto Cut-Outs: High-resolution (8000×8000 pixel) orthophoto images for each monitored intersection, used for georeferencing and visualization.

Road and Lane Segmentations: CSV files defining lane polygons within road sections, facilitating mapping of vehicle positions to road segments and lanes.

Sample Videos: A selection of 4K UHD drone video samples capturing intersection footage during the experiment.

Data Collection

The dataset was collected as part of a collaborative multi-drone experiment conducted by KAIST and EPFL in Songdo, South Korea, from October 4–7, 2022.

A fleet of 10 drones monitored 20 busy intersections, executing advanced flight plans to optimize coverage.

4K (3840×2160) RGB video footage was recorded at 29.97 FPS from altitudes of 140–150 meters.

Each drone flew 10 sessions per day, covering peak morning and afternoon periods.

The experiment resulted in 12TB of 4K raw video data.

More details on the experimental setup and data processing pipeline are available in [1].

Data Processing

The trajectories were extracted using geo-trax, an advanced deep learning framework designed for high-altitude UAV-based traffic monitoring. This state-of-the-art pipeline integrates vehicle detection, tracking, trajectory stabilization, and georeferencing to extract high-accuracy traffic data from drone footage.

Key Processing Steps:

Vehicle Detection & Tracking: Vehicles were detected and tracked across frames using a deep learning-based detector and motion-model-based tracking algorithm.

Trajectory Stabilization: A novel track stabilization method was applied using detected vehicle bounding boxes as exclusion masks in image registration.

Georeferencing & Coordinate Transformation: Each trajectory was transformed into global (WGS84), local Cartesian, and orthophoto coordinate systems.

Vehicle Metadata Estimation: In addition to time-stamped vehicle trajectories, various metadata attributes were also extracted, including vehicle dimensions and type, speed, acceleration, class, lane number, road section, and visibility status.

More details on the extraction methodology are available in [1].

File Structure & Formats

1. Trajectory Data (Daily Intersection ZIPs, 16.2 MB ~ 360.2 MB)

The trajectory data is organized into 80 ZIP files, each containing traffic data for a specific intersection and day of the experiment.

File Naming Convention:

YYYY-MM-DD_intersectionID.zip

YYYY-MM-DD represents the date of data collection (2022-10-04 to 2022-10-07).

intersectionID is a unique identifier for one of the 20 intersections where data was collected (A, B, C, E, …, U). The letter D is reserved to denote "Drone".

Each ZIP file contains 10 CSV files, each corresponding to an individual flight session:

YYYY-MM-DD_intersectionID.zip
│── YYYY-MM-DD_intersectionID_AM1.csv
├── …
│── YYYY-MM-DD_intersectionID_AM5.csv
│── YYYY-MM-DD_intersectionID_PM1.csv
├── …
└── YYYY-MM-DD_intersectionID_PM5.csv

Here, AM1-AM5 and PM1-PM5 denote morning and afternoon flight sessions, respectively. For example, 2022-10-04_S_AM1.csv contains all extracted trajectories from the first morning session of the first day at the intersection 'S'.

CSV File Example Structure:

Each CSV file contains high-frequency trajectory data, formatted as follows (d.p. = decimal place):

Dataset Column Name
Format / Units
Data Type
Explanation

Vehicle_ID
1, 2, …
Integer
Unique vehicle identifier within each CSV file

Local_Time
hh:mm:ss.sss
String
Local Korean time (GMT+9) in ISO 8601 format

Drone_ID
1, 2, …, 10
Integer
Unique identifier for the drone capturing the data

Ortho_X, Ortho_Y
px (1 d.p.)
Float
Vehicle center coordinates in the orthophoto cut-out image

Local_X, Local_Y
m (2 d.p.)
Float
KGD2002 / Central Belt 2010 planar coordinates (EPSG:5186)

Latitude, Longitude
° DD (7 d.p.)
Float
WGS84 geographic coordinates in decimal degrees (EPSG:4326)

Vehicle_Length*, Vehicle_Width*
m (2 d.p.)
Float
Estimated physical dimensions of the vehicle

Vehicle_Class
Categorical (0–3)
Integer
Vehicle type: 0 (car/van), 1 (bus), 2 (truck), 3 (motorcycle)

Vehicle_Speed*
km/h (1 d.p.)
Float
Estimated speed computed from trajectory data using Gaussian smoothing

Vehicle_Acceleration*
m/s² (2 d.p.)
Float
Estimated acceleration derived from smoothed speed values

Road_Section*
N_G
String
Road section identifier (N = node, G = lane group)

Lane_Number*
1, 2, …
Integer
Lane position (1 = leftmost lane in the direction of travel)

Visibility
0/1
Boolean
1 = fully visible, 0 = partially visible in the camera frame

* These columns may be empty under certain conditions, see [1] for more details.

2. Orthophoto Cut-Outs (orthophotos.zip, 1.8 GB)

For each intersection, we provide the high-resolution orthophoto cut-outs that were used for georeferencing. These 8000×8000 pixel PNG images cover specific areas, allowing users to overlay orthophoto trajectories within the road network.

orthophotos/
│── A.png
│── B.png
│── …
└── U.png

For more details on the orthophoto generation process, refer to [1].

3. Orthophoto Segmentations (segmentations.zip, 24.9 KB)

We provide the road and lane segmentations for each orthophoto cut-out, stored as CSV files where each row defines a lane polygon within a road section.

Each section (N_G) groups lanes moving in the same direction, with lanes numbered sequentially from the innermost outward. The CSV files are structured as follows:

segmentations/
│── A.csv
│── B.csv
│── …
└── U.csv

Each file contains the following columns:

Section: Road section ID (N_G format).

Lane: Lane number within the section.

tlx, tly, blx, bly, brx, bry, trx, try: Polygon corner coordinates.

These segmentations enabled trajectory points to be mapped to specific lanes and sections in our trajectory dataset. Vehicles outside segmented areas (e.g., intersection centers) remain unlabeled. Perspective distortions may also cause misalignments for taller vehicles.

4. Sample Videos (sample_videos.zip, 26.8 GB)

The dataset includes 29 video samples, each capturing the first 60 seconds of drone hovering over its designated intersection during the final session (PM5) on October 7, 2022. These high-resolution 4K videos provide additional context for trajectory analysis and visualization, complementing the orthophoto cut-outs and segmentations.

sample_videos/
│── A_D1_2022-10-07_PM5_60s.mp4
│── A_D2_2022-10-07_PM5_60s.mp4
│── B_D1_2022-10-07_PM5_60s.mp4
│── …
└── U_D10_2022-10-07_PM5_60s.mp4

Additional Files

README.md – Dataset documentation (this file)

LICENSE.txt – Creative Commons Attribution 4.0 License

Known Dataset Artifacts and Limitations

While this dataset is designed for high accuracy, users should be aware of the following known artifacts and limitations:

Trajectory Fragmentation: Trajectories may be fragmented for motorcycles in complex road infrastructure scenarios (pedestrian crossings, bicycle lanes, traffic signals) and for certain underrepresented truck variants. Additional fragmentations occurred when drones experienced technical issues during hovering, necessitating mid-recording splits that naturally resulted in divided trajectories.

Vehicle ID Ambiguities: The largest Vehicle_ID in a CSV file does not necessarily indicate the total number of unique vehicles.

Kinematic Estimation Limitations: Speed and acceleration values are derived from raw tracking data and may be affected by minor errors due to detection inaccuracies, stabilization artifacts, and applied interpolation and smoothing techniques.

Vehicle Dimension Estimation: Estimates may be unreliable for stationary or non-axially moving vehicles and can be affected by bounding box overestimations capturing protruding vehicle parts or shadows.

Lane and Section Assignment Inaccuracies: Perspective effects may cause vehicles with significant heights, such as trucks or buses, to be misassigned to incorrect lanes or sections in the orthophoto.

Occasional pedestrian pair misclassifications: Rarely, two pedestrians walking side by side may be briefly mistaken for a motorcycle, but such instances are short-lived and typically removed by the short trajectory filter.

For a comprehensive discussion of dataset limitations and validation procedures, refer to [1].

Citation & Attribution

Preferred Citation:

If you use Songdo Traffic for any purpose, whether in academic research, commercial applications, open-source projects, or benchmarking efforts, please cite our accompanying manuscript [1]:

Robert Fonod, Haechan Cho, Hwasoo Yeo, Nikolas Geroliminis (2025). Advanced computer vision for extracting georeferenced vehicle trajectories from drone imagery, arXiv preprint arXiv:2411.02136.

(Note: This manuscript shall be replaced by the published version once available.)

Note: Although Zenodo automatically provides a formal dataset citation (shown below), we kindly request that you reference the manuscript as the primary source of this work.

Dataset Citation (for archival purposes):

Robert Fonod, Haechan Cho, Hwasoo Yeo, Nikolas Geroliminis (2025). Songdo Traffic: High Accuracy Georeferenced Vehicle Trajectories from a Large-Scale Study in a Smart City (v1). Zenodo. DOI: 10.5281/zenodo.13828384.