Aerial Video & Trajectory Dataset Of Vehicles On Circular Road

This dataset presents aerial video and vehicle trajectory data collected during a phantom traffic jam experiment with the Swiss television (SRF) at the driving test center TCS Derendingen, Solothurn, from March 12th 2024. 14 vehicles were recorded for a total duration of 40 minutes with a drone from above, and vehicle trajectories were extracted using computer vision and Kalman filtering methodology.
The observed vehicles differ by their power train (combustion, electric, hybrid), gearbox (manual, automatic), and equipment with advanced driver assistance systems. 

The folder VIDEOS contains fifteen video recordings covering a duration of around one hour (01:02:56) and 94,423 frames in total, at a framerate of 25 frames per second, and with a resolution of 3840 x 2160 pixels. The videos show more than 14 vehicles driving on the ring road of the driving test center TCS Derendingen (Solothurn, Switzerland) during various experiments. The videos are provided in MOV format.

The folder ANNOTATIONS provides object annotations for each video and frame a list of rectangular annotations that envelop a vehicle, generated by 18 different object detection models.
The annotations are provided as zipped CSV files, separated by the tabulator symbol.
Each row consists of eight columns: 

• (1) frame number
• (2) annotation type (following the DOTA dataset classification system)
• (3) rectangle center x coordinate (in pixels)
• (4) rectangle center y coordinate (in pixels)
• (5) rectangle width (in pixels)
• (6) rectangle height (in pixels)
• (7) rectangle angle (in radians)
• (8) annotation confidence (between 0.0 and 1.0, provided by the neural network).

The folder TRAJECTORIES contains for each video, vehicle, and frame an exact vehicle position, speed, acceleration, and headway. The vehicle trajectories are provided as zipped CSV files, separated by the comma symbol.
Each row consists of 18 columns:

• (1) Vehicle_ID 
• (2) Frame_ID
• (3) Global_Time (in seconds)
• (4) Cartesian_X (in meter)
• (5) Cartesian_Y (in meter)
• (6) Polar_X  (in radians)
• (7) Polar_Y  (in meter)
• (8) Lane_X  (in meter)
• (9) Lane_Y  (in meter)
• (10) v_Length  (in meter)
• (11) v_Width  (in meter)
• (12) v_Vel  (in m/s)
• (13) v_Angle (in radians)
• (14) v_AngleVel (in rad/s)
• (15) Proceeding (Vehicle_ID of proceeding vehicle)
• (16) Space_Hdwy (in meter)
• (17) Time_Hdwy (in seconds)
• (18) v_Accel (in m/s^2)

The code to extract and process the vehicle trajectories from the videos is available on GitHub: https://github.com/DerKevinRiehl/trajectory_analysis 

The TV-show episode of "Einstein" in Swiss television "SRF" from May 2nd 2024 can be found here: https://www.srf.ch/play/tv/einstein/video/stau-was-hilft-gegen-den-verkehrskollaps?urn=urn:srf:video:63965781-c7ea-4033-9827-be4275f1cba5

Acknowledgements

We thank the Schweizer Radio und Fernsehen (SRF, Swiss Radio and Television, tv-show "Einstein"), Adrian Winkler, Laurin Merz, and Andrea Fischli for their support when organizing participants and vehicles for the experiment, and filming and documenting it for the Swiss public. We thank Andre Greif and the TCS Driver Training Center in Derendingen, Solothurn (Switzerland) for hosting our experiment.

Publications

• "Aerial Video & Trajectory Dataset Of Vehicles On Circular Road" (In Submission in Data In Briefs)
• "Consistent Vehicle Trajectory Extraction From Aerial Recordings Using Oriented Object Detection" (In Submission in Scientific Reports)