Traffic Data

Vehicle trajectory datasets worth knowing: what each one shows, and how to get it

A trajectory dataset records the position of every vehicle, many times a second. It is the richest traffic data there is: from it you can rebuild what a loop detector, a probe vehicle or a drone would have seen (Lecture 2). The datasets below are open to researchers. Each has its own terms of use, so read them before you download, and cite the paper the authors ask for.

NGSIM

  • What: vehicle trajectories at 10 Hz, extracted from video recorded from tall buildings in 2005–2006, on freeway sections of US-101 and I-80 in California and on two urban arterials.
  • Good for: time-space diagrams, congestion and stop-and-go waves, car-following. It is the most widely used trajectory dataset in traffic research.
  • Know its flaws: the raw data contain position noise and some impossible speeds and accelerations. Coifman and Li have since re-extracted the I-80 trajectories from the original video to fix them.
  • Access: free, from the U.S. DOT open-data portal. Lab 1 downloads it for you.
  • Cite: U.S. Department of Transportation, Federal Highway Administration (2016). Next Generation Simulation (NGSIM) Vehicle Trajectories and Supporting Data. https://doi.org/10.21949/1504477

I-24 MOTION

  • What: 276 pole-mounted cameras cover about 4.2 miles of I-24 near Nashville, Tennessee, a freeway with four to five lanes in each direction and frequent congestion. Computer vision turns the video into trajectories of every vehicle, at 25 Hz, with its length, width and class.
  • Released so far: the first dataset, INCEPTION (2023), has at least four hours of traffic on each of ten days. The team plans to release more as the system matures.
  • Good for: stop-and-go waves along miles of freeway, not a few hundred metres; the same stretch day after day.
  • Access: free to download from i24motion.org, for academic and commercial work, under a data-use agreement.
  • Cite: Gloudemans, D., Wang, Y., Ji, J., Zachar, G., Barbour, W., Hall, E., Cebelak, M., Smith, L. and Work, D. B. (2023). I-24 MOTION: An instrument for freeway traffic science. Transportation Research Part C 155, 104311. https://doi.org/10.1016/j.trc.2023.104311

highD

  • What: drone video of German highways at six locations, turned into trajectories of about 110,500 cars and trucks at 25 Hz, with positions typically accurate to within 10 cm. Each vehicle comes with its size, type and lane changes.
  • Good for: car-following and lane changing at highway speeds, and clean data with little noise. Most of it is free-flowing traffic, and only a few recordings show jams.
  • Access: free for non-commercial research, through an application form at levelxdata.com. Each request is reviewed, and the data may not be passed on, so each user applies for their own copy.
  • Cite: Krajewski, R., Bock, J., Kloeker, L. and Eckstein, L. (2018). The highD dataset: A drone dataset of naturalistic vehicle trajectories on German highways for validation of highly automated driving systems. IEEE International Conference on Intelligent Transportation Systems (ITSC). https://doi.org/10.1109/ITSC.2018.8569552

exiD

  • What: from the same team as highD: drone video of highway entries and exits at seven locations in Germany, about 69,000 road users, with a road map of each site in OpenDRIVE and Lanelet2 formats.
  • Good for: merging, cut-ins and lane changes near ramps, where drivers interact the most. It is weaker for stop-and-go waves, since most of it is free-flowing.
  • Access: same terms as highD: free for non-commercial research and teaching, on application at levelxdata.com, and no redistribution.
  • Cite: Moers, T., Vater, L., Krajewski, R., Bock, J., Zlocki, A. and Eckstein, L. (2022). The exiD dataset: A real-world trajectory dataset of highly interactive highway scenarios in Germany. IEEE Intelligent Vehicles Symposium (IV). https://doi.org/10.1109/IV51971.2022.9827305

CitySim

  • What: drone video at 30 frames a second, 1,140 minutes in all, turned into vehicle trajectories with rotated bounding boxes. Sites in the U.S. include signalized and unsignalized intersections and one- and two-lane roundabouts; partner teams in China and Hong Kong host freeway, weaving and merge sites.
  • Good for: safety studies with surrogate measures such as time-to-collision, and building digital twins of real sites.
  • Access: from the CitySim GitHub page, University of Central Florida. The partner sites are requested from their own teams.
  • Cite: Zheng, O., Abdel-Aty, M., Yue, L., Abdelraouf, A., Wang, Z. and Mahmoud, N. (2023). CitySim: A drone-based vehicle trajectory dataset for safety-oriented research and digital twins. Transportation Research Record. https://doi.org/10.1177/03611981231185768

SinD

  • What: drone recordings of signalized intersections in China, with the traffic-light state, an HD map, and trajectories of seven types of road user, from cars to cyclists and pedestrians. The first release covers one intersection in Tianjin: about 7 hours and more than 13,000 road users, most of them vulnerable road users such as pedestrians and cyclists. A 2026 update, SinD 2.0, extends it to six intersections in four cities, with labelled safety-critical events.
  • Good for: intersections with the signal state known, red-light running, and interactions between cars and vulnerable road users.
  • Access: free for non-commercial research; apply by email with an institutional address, as described on the SinD GitHub page.
  • Cite: Xu, Y., Shao, W., Li, J., Yang, K., Wang, W., Huang, H., Lv, C. and Wang, H. (2022). SIND: A drone dataset at signalized intersection in China. IEEE International Conference on Intelligent Transportation Systems (ITSC). https://arxiv.org/abs/2209.02297

TUMTraf-A

  • What: ten real crashes on a German autobahn, recorded by roadside cameras and LiDAR on the A9 test bed near Munich, with 3-D boxes and track IDs for every road user. Crashes are rarely caught on sensors; here each one is recorded from several views.
  • Good for: what happens in the seconds before and after a crash, and how well automatic crash detection works.
  • Access: under a CC BY-NC-SA 4.0 licence (non-commercial), from the TUMTraf-A page, Technical University of Munich.
  • Cite: Zimmer, W., Greer, R., Zhou, X., Song, R., Pavel, M., Lehmberg, D., Ghita, A., Gopalkrishnan, A., Trivedi, M. and Knoll, A. (2025). Safety-critical learning for long-tail events: The TUM Traffic Accident dataset. https://arxiv.org/abs/2508.14567
NoteWhich one for what?
  • Freeway waves and car-following: NGSIM, I-24 MOTION
  • Lane changes and merges: highD, exiD
  • Intersections and roundabouts: CitySim, SinD
  • Pedestrians and cyclists: SinD
  • Crashes and near misses: TUMTraf-A for crashes; CitySim and SinD for near misses