Another new MapRoulette challenge: Suspected Missing / Wrong One Way Streets

Posted by mvexel on 11 April 2014 in English (English)

Over at Telenav we have been busy comparing the GPS traces of drivers using our navigation apps against OSM data. This is fun work and sometimes leads to really useful results. One of our recent studies involved comparing what we think are one-way streets to the corresponding OSM ways. The process is quite simple: count how many traces following a given street go in one direction, and how many go in the other. If the number for one direction is close to zero, the street is probably one-way. The outcome depends on the confidence level you apply, but it turns out that there are more than 100 thousand OSM ways that probably should be marked with a oneway tag, but aren't.

One. Hundred. Thousand.

I don't know about you, but I think that would make for a really great MapRoulette challenge. So that's what I did, and here it is.

But how do you know if a street should be one-way without having eyes on the ground? It turns out the aerial imagery provides helpful hints. Let me show you what I look at to verify the one-way-ness of a street using just JOSM and the Bing layer.

Let's look at this case here:

Notice the arrowhead? That is the direction we found the vast majority of trips followed, so we think that is the one-way direction of this street. Let's look in JOSM:

In this case, there is really only one hint that this is in fact a one way street, but it's a pretty convincing one: The markings on the road. See the stop line? It goes all the way across, which it wouldn't if traffic were allowed to enter from the right. Also, it is a little blurry, but there is what appears to be a left arrow painted on the road as well. So all in all, I am convinced that this way should be marked oneway=yes.

Be careful though! OSM ways can be very long, and the one-way restriction may not apply to the entire way. So always be sure to check the entire length of the way before you tag and upload, and split the way as necessary! The Telenav detection works on smaller segments internally, so the suggested one-way-ness does not always apply to the entire OSM way.

Lets' look at another case.

Looks like a residential area. Let's load it up in JOSM:

In this case, it's really hard to see. There's no obvious stopping line that I could see, and there's too much shade to see if there are parked cars on the left side of the street that are parked facing west - which would be another good indication that it is a one-way street, because it is illegal in the United States to park in the opposite direction of the flow of traffic. So this one I would mark as 'Couldn't See'. Perhaps someone else sees something I don't? Here is the permalink to the task in MapRoulette. (Neat eh, permalinks - makes it easy to share a task you want to discuss or highlight. They remain valid even after a task is fixed or even deleted.)

Oh and yes, I am afraid this is another United States only challenge... We are working hard to get more international challenges in, and we already have some interest. I am sure I mentioned this before, but MapRoulette now has an API that lets 'third parties' submit their own challenges. On top of that, we're also making it really easy to deploy your own MapRoulette server. If you're interested in any of that, just let me know. I will be hosting a session on MapRoulette at State Of The Map US this weekend, and Serge and are also planning to run a BoF session for a deeper dive into MapRoulette challenge making. If you don't make it to SOTM US, don't feel left out - there will likely also be MapRoulette sessions at SOTM EU in June!

Happy Mapping!

Comment from Mappo on 11 April 2014 at 08:12

That's a really cool use of data to improve the map. But because it might need on-site verification perhaps it should also be fed into one of the QA tools that lets you scan your local area for (potential) problems too.

Comment from robbieonsea on 11 April 2014 at 10:11

Are you using Telenav GPS traces (so presumably only driving traces) or OSM traces?

Otherwise for OSM traces one would need to filter out traces from walking which presumably can go in any direction, for example perhaps just discarding traces with slow average speeds when performing vehicular type analysis.

Can you also enlighten us on the 'confidence level' methodology? (or mention in your SOTM talks)

Thanks for all the improvements to MapRoulette.

Comment from Chris Davis on 14 April 2014 at 13:01

Amusingly, I came across this diary entry of yours because you incorrectly changed a section of street in my neighborhood to one way. (changeset 21194602). The street was changed to two way 18 months ago, so you are working from old aerial photos.

I do realize that the vast majority of streets you are changing to one-way aren't going to have actually been changed in the last year or two, but it's something to keep in mind.

Comment from mvexel on 15 April 2014 at 13:34

Chris: That is a very good point, and I should (and will!) include a note of caution to MapRoulette users. I am sorry I messed up data in your area, and thanks for watching over it.

robbieonsea: These are Telenav traces only - so yes, these are driving traces, not walking or cycling.

The confidence level is a simple ratio of traces going in one direction versus traces going in the other, and for the results that are in this particular MapRoulette challenge, we set the threshold at 95%, meaning that for a way to be flagged we will see at least 95 out of 100 traces going in one direction for that segment. There is also a minimum threshold for the number of traces we want to see, but I cannot recall what that threshold is.

Incidentally, what you see right now is the first batch of 5000 tasks out of a total of over a hundred thousand, and I will load them sorted by the number of traces. So the tasks you are seeing in MapRoulette today are the ones that have the most traces attached to them.

Ideally, as Mappo also mentioned, this challenge would be fed to tools that allow for on site verification. Osmose is a good candidate, especially now that Frederic is doing a ton of work to make that tool mobile-friendly. I have been working with Frederic over the SOTM US weekend to build a bridge between Osmose and MapRoulette, so we will see an exchange of tasks / challenges between these two QA tools in the near future.

Thanks everyone for the feedback!

Comment from rickmastfan67 on 19 April 2014 at 03:15

which would be another good indication that it is a one-way street, because it is illegal in the United States to park in the opposite direction of the flow of traffic.

If this was true Martin, a ton of my neighbors should have gotten tons of tickets. I live on a two-way road and they park facing the opposite direction on each side of the road almost every day. And we even have a few cops that live in the neighborhood.

Login to leave a comment