Hi guys, I am the creator of this dataset. Together with Allister, I am happy to help you out if you have any questions.

Yes, it is cars only and mostly European cities, and yes that's a limitation. Nonetheless, it has advantages compared to GPS/Bluetooth data, as it captures ALL cars (or trucks). If you're interested in car capacity of a network, macro or micro congestion, or other applications that require rigorous detection of ALL vehicles, this dataset should get you started! We're happy to hear from you!

Lukas

Just wondering if there's a full list of cities somewhere? I only see a carousel and a (non-clickable) map.

I see a city in the middle of the US but I can't make out what it is.

EDIT: Oh it's Toronto. I couldn't localize without the Great Lakes on the map.

very interesting data! Would be even moreso if you were able to get data from a city with a large population like Lagos or Delhi

Thanks for your interest. Tokyo, Taipeh, Los Angeles, Paris, and London are covered :)

No Rome, no Athens, no Istanbul, no Bucharest, in part I understand that but collecting data from the “traffic-civilized” part of the continent will only tell a partial story. Maybe from the included list Paris comes close to those cities, even though I doubt it.

When you're poor you buy cheap ground beef. When you're rich buying $8/lb grass fed fancy beef seems like reasonable use of money. Governments are the same way. Only "rich Europe" can afford/justify carpet bombing their infrastructure with sensors so they're the only ones the data covers.

You see this incidental data bias in lots of fields. "Corrupt seaside Europe", "dashcam videos and good vodka Europe" and "I Can't Believe It's Not The Middle East(TM) Europe have bigger more immediate fish to spend their dollars frying than collecting data for the sake of maybe finding some utility in it later on.

I agree with you, the issue is that the technocratic powers from those cities I mentioned (I for myself live in Bucharest) will most definitely use the studies made up using this type of data and will apply them as they are, using “the Germans/Swiss/Dutch are doing it so we should, too” as their main reasoning. By the time someone comes in and says that most probably the situations between Bucharest and Amsterdam (to give a random example) are very different it is too late, the measures have already been taken.

If the US is any indication it will be worse than that. There are all sorts of topics in which "what works in Europe" factors into the discourse yet despite the topic being something on which most of the world has kept good records on for 50+yr north/central/west Europe is still the only part that is paid any attention to.

If you're interested in Athens, I warmly recommend our good friends' work from EPFL: https://open-traffic.epfl.ch/ Traffic data from drones for Athens.

cool open data! (I used it as an answer here: https://opendata.stackexchange.com/q/1767/1511)

for other traffic related datasets, check out this list: https://github.com/graphhopper/open-traffic-collection

and for "borrowing" the live data from TomTom, check out this mini tweet thread (from me): https://twitter.com/philshem/status/1241739025624567813

thanks for the efforts! I have added a link to your github on our github site. Cheers, Lukas

In total, almost 5 billion vehicles covering a combined time span of 3.8 years were detected. The UTD19 traffic data is free for all research use. You only have to sign up, agree with the conditions, and you are all set.

We are now in the age of multimodal mobility and this only covers the mode that is easiest to capture (vehicles using data from loop detectors). Having a dataset with the same spatial and temporal coverage but multimodal would be amazing.

TBH, this is already quite a step. There's much more data available. I worked on systems that also had LPRs (license plate readers), and they generate more detailed information: you'd get 1M vehicle passings per day in a medium-sized city. But they're privacy sensitive. Same goes for bluetooth detection or face recognition. Data owners aren't going to expose that kind of fine-grained data to the general public easily.

Do you know if there are datasets with pseudonymized LPR data available? I'm not interested in the actual license plate obviously, but being able to so say that two data points are the same car would be extremely interesting data especially for these larger and larger datasets spanning longer periods of time.

I'm also aware there is still a bit of a privacy concern with that type of data but honestly don't really have the math background to know exactly how that occurs or the extent to which you can minimize it.

Riding a bike I'm still interested in the volume of cars on the road, and probably more interested in the volume of cars than the volume of bikes to be honest.

It depends on the needs. In your case (it seems) you are looking for the safest (on shared infrastructure) and / or cleanest routes for biking. And the same value can be extracted for vehicle users (more efficient routes). In my case, it is mobility planning. I wasn't denying the value of the dataset, I was just asking for more :)

I recently learned there are a few groups at ETZH working on traffic simulation and fleet planning in transportation networks. Are you affiliated with the Autonomous Mobility on Demand / IDSC folks?

https://www.amodeus.science/

https://idsc.ethz.ch/education/lectures/duckietown.html

Yes, they're amazing! We have collaborations with them, but we are from: https://www.ivt.ethz.ch/ focusing more on the transportation/traffic part of research. Thanks for the interest!

the site is currently down, as ETH is experiencing issues with their webhosting. Will keep you posted.

EDIT: Up again!