Hi, we are TomTom - let's exchange ideas

No, the mail adres on your page Organised Editing page.

Apparently we missed that message, I’m sorry about that. I will shoot you an email.

For your information: the new MapMetrics documentation I was talking about before is now available in the help section of MapMetrics.

Kind regards,
Marjan

At this time we are not using OSM data in any of our commercial services. If this changes, we will abide by the OSM community guidelines.

Eens doorgelezen maar ik begrijp dat gereden routes worden vergeleken met dezelfde route maar dan door Grasshopper berekend op basis van OSM data. Hoe meer de gereden routes afwijken van de route van Grasshopper hoe roder de tile wordt.

Eens in de data bij mij in de beurt gekeken maar een patroon kan ik er niet in vinden en al helemaal niet aanwijzingen waar de kaart verbeteringen nodig heeft. Wat dat betreft is Scheve_prio1_wegen_QGC kaart van @PeeWee32 heel veel meer bruikbaar, zie ook Waar ligt OSM highway het meeste scheef?

Thank you for sharing this. As @emvee pointed out and as listed in the documentation, GraphHopper was used to match the TomTom GPS traces to the road network. I can imagine some of the false positives stem from this process, especially given unorthodox movements some carriers of TomTom GPS logging devices may exhibit (e.g. movement against driving direction). A huge step would be if TomTom open sourced (part of) these anonymised GPS traces to the (research) community, such that we could run our own map matching algorithms and routers against the data and come up with even more valuable insights into the quality of the OSM road network (I would personally be very interested in this since I have done map matching experiments against the OSM road network in a research setting in the past). Perhaps it would even be possible to launch a Kaggle challenge with a subset of the TomTom GPS traces, divided into a test and train set, in order for the larger community outside TomTom to come up with the best possible way of using this data to find potential improvements in OSM. Would love to hear your thoughts on this!

Yes, make a Tomtom heatmap, just as we now have from Strava. For now Mapmetrics is unusable. But even then, here in the Netherlands we have much more options with different layers like BGT.

Hi, we will pass along that feedback to Steve C. and his team. Thank you!

MapMetrics has issues we’ve been working on. Mainly this is trainlines and water, as I said in the blog post about it. However it is a moving target, later months are better and we’ve been going back and rerunning prior months.

The existential issue is we can’t share the GPS data with you. In rural areas, it’s usually very obvious what’s wrong even without GPS. In denser areas it’s not so obvious (but can be with GPS). Sometimes it’s more hidden things like turn restrictions, or lots of people breaking the law (going down one way streets the wrong way for examples). But in any case, we’re usually pretty confident about “red” cells. If you have a particular case I can look at it, send a MM link…

Can’t → won’t. You own that data don’t you?

I had a look using OSM202107 and found on the A13 in Rotterdam a tile that is (dark) red according to the .pdf, the Median Count is larger than 50000:

Would be good if you can do a deep dive on this so it is clear why according to MapMetrics there are so many problems in this tile because I would be really surprised if you can find a real problem in the OSM data.

Could you show us the GPS tracks backing the cell emvee shared above?

I would also like to see these:

At first glance there seem to be rather a lot of false positives.

Also look at what happens on aerodromes; something is going wrong there too.

Earlier I mentioned the situation on the Markerwaarddijk.

http://mijndev.openstreetmap.nl/~allroads/NLOSMOK/OSMOKNL.html#map=18.20/52.6693/5.3682/0

What about a cyclist and a tomtom app on his smarpthone? Are the Markerwaarddijk false positive come from this use?

Ride with a car, stop on a parking, cycle further, still the application active. Are the cyclist traces filtered?

I know people riding with their smartphone in their pocket and use a headset.

If Tomtom is going to map and tag, how to correct problems, survey, data made and used bij Tomtom is like as our own survey and images taken of the situation. Their streetview imagery.

Can we use the survey of Tomtom, their streetview imagery to solve problems?

When is the moment that the licence should be checked when used for Openstreetmap?

Ok so for this one

https://mapmetrics-world.innovation.tomtom-global.com/#?mapSource=OSM&metric=median_count&z=18&lat=51.939092&lon=4.429593&ds=OSM202107&ds2=None&mapSource2=OSM

Here’s what I found in no particular order:

  • There are mopeds (I think) using Kleinpolderkade but not super many of them. We do filter cyclists with some success. Mopeds are harder :-/
  • The car park in Piet Cottarstraat has many GPS traces going through the middle of it, right down the middle of where satellite shows the car park recently. It looks like either some GPS is snapped to a map with a road through the middle (which is possible but unlikely from looking at the maps), or, there was road work and it was redirected.
  • The one way streets in OSM on the West side of the A13 don’t agree with multiple other map providers. No idea who is right without being on the ground.
  • There’s a chance that the two living streets from the car park to Ludolf de Jonghstraat aren’t being routed against correctly, we will check.

Zooming around one of the links provided:

For here:

https://mapmetrics-world.innovation.tomtom-global.com/#?mapSource=OSM&metric=median_count&z=18&lat=53.182382&lon=5.758265&ds=OSM202107&ds2=None&mapSource2=OSM

That building(?) at the SE corner of the roundabout is full of vehicle traces. On the satellite I looked at it looks like a rough grass square. From the GPS it looks like a car park or loading docks.

Here’s another

https://mapmetrics-world.innovation.tomtom-global.com/#?mapSource=OSM&metric=median_count&z=18&lat=53.169840&lon=5.724118&ds=OSM202107&ds2=None&mapSource2=OSM

OSM has a cyclepath with a access gate to the east of what looks like a house. From GPS, this looks like a lot of cars with no gate and should “really” be a road. It could be moped riders but looks like cars. It doesn’t look like anyone stops for the gate, if it’s there.

From another link, looking at the red areas here:

https://mapmetrics-world.innovation.tomtom-global.com/#?mapSource=OSM&metric=median_count&z=18&lat=53.222452&lon=5.766744&ds=OSM202107&ds2=None&mapSource2=OSM

If you look at satellite there’s lots of access roads missing around the buildings and connecting to what looks like a taxiway (which I can see a lot of vehicles use).

Similarly:

https://mapmetrics-world.innovation.tomtom-global.com/#?mapSource=OSM&metric=median_count&z=18&lat=53.223758&lon=5.696361&ds=OSM202107&ds2=None&mapSource2=OSM

has roads to the northern buildings, (road or highway:track, or something but there’s no road there in OSM)

For this one:

https://mapmetrics-world.innovation.tomtom-global.com/#?mapSource=OSM&metric=median_count&z=18&lat=53.191607&lon=5.633461&ds=OSM202107&ds2=None&mapSource2=OSM

the “space” in that tile at the top right that runs roughly NW/SE is a heavily trafficed “road” from the GPS.

On the other hand, this one is a complete puzzle to me:

https://mapmetrics-world.innovation.tomtom-global.com/#?mapSource=SAT&metric=median_count&z=17&lat=53.190842&lon=5.632614&ds=OSM202107&ds2=None&mapSource2=OSM

This is also another slightly puzzling one:

https://mapmetrics-world.innovation.tomtom-global.com/#?mapSource=SAT&metric=median_count&z=18&lat=53.171228&lon=5.60574&ds=OSM202109&ds2=None&mapSource2=OSM

It looks to me like a lot of people take mopeds (or something) past that building in this red tile. Filtering out high speed cyclists vs. low speed mopeds is not trivial.

Hope this was useful we will discuss these more internally and see if we can figure out solutions. More feedback is welcome!

I’m not sure which gate you are referring to, but you are right that there is a tagging flaw there. The cycleway doesn’t start until just south of that farm. I’ve fixed it.

Your notes are interesting, thanks for taking the time to look into these. However, without seeing the (compound) GPS traces for cells ourselves we simply can’t provide any meaningful feedback on why some cells are red and some are fine except in a very limited piece-meal fashion.

As an analytical tool I find it, in its current state, just not very useful for the average mapper.

Could you explain why you are holding out on showing us the traces backing these in a form that is amenable to your apparent need for secrecy? Even discussing this selection of cells you’re not showing your source data leaving us to just imagine them. I’m sure the raw GPS traces themselves are an asset you are forbidden to share with competitors by your superiors, but surely you can generate a sort of heat map to complement your MapMetrics cells?

As I provided this tile I did have a look.

  • The Kleinpolderkade was not part of the tile but reviewing it I saw it was mapped as highway=pedestrian but that is not correct, there is no G11 and cars can drive there provided the bollard at the begin/end are removed. For sure mopeds can drive there, but it is not a logical through route.

  • The buildings of the Piet Cottarstraat were finished 2019 so road work looks unlikely to me

  • I reviewed the one way streets in OSM on the West side of the A13 (als not part of the tile) using the verkeersbordenkaart and they were all correct apart from one piece of the Julianastreet that I corrected. What are your “multiple other map providers”?

  • The two living streets between the car park to Ludolf de Jonghstraat and the Ludolf de Jonghstraat are not living streets but pedestrian area’s that can also be used by bicycles/moped’s. Clearly visible on PDOK/BGT/Obligue so corrected that. I do not expect any car traffic on these ways.

A few more things:

  • Please look at the latest dates. On the layer panel on the right you should see 07 (july), 08, 09. 10 is coming. We had some issues with 09 tiles we’re fixing now. You should notice that the detection gets better over time, especially with railways.
  • We may actually just remove old months if you folks find the old ones confusing or not useful, and just always show the latest data
  • Any feedback on the comparison (month-to-month) layers?
  • We’re working on figuring out cyclists but it will take time

Generally speaking I think the “red tiles” are pretty good with the caveats about cyclists and (in the past) trains. It’s unfortunate we can’t open up GPS but there it is. Red tiles with one road in them are generally trivial to see what the problem is (as you can see in some of my prior examples). Red tiles with 20 roads need some sleuthing but generally are still pretty good. We can make smaller tiles to give a higher resolution grid, we’ll investigate this.

  • “multiple other map providers” - I also looked at google and tomtom I think, purely to see if they agreed or not.
  • cycle lanes - we route right now for cars. So if mopeds or cyclists get in the data, and use cycle paths, which they do a lot in NL, then we have problems. As I mentioned, we are working on a variety of solutions for this.
  • GPS - it’s primarily privacy reasons and unfortunately a heatmap is not a sufficient obfuscator.

Maybe you can show part of the GPS trace in the red tiles used to indicate the error and leave all traces without issue away and remove GPS trace as soon as the issue is resolved?
Or at least indicate with a pin with some text (like osmose or keepright) what’s wrong. Makes it more useful and likely that someone will fix it and less like looking for a needle in the haystack…

I can’t say that I see how a low resolution heat map of GPS traces — i.e., all GPS traffic from your data source compounded into a single image corresponding to your cells, similar to what Strava had — is a privacy concern for anyone as long as you exclude conflict zones (of which we obviously have none within our national borders). Could you explain this? Perhaps you need to talk to your privacy officer about this to obtain permission? You can reach them via privacy@tomtom.com.

We’re not asking for raw data linked to user accounts, we want to see the compound data to compare it with the mapped ways in order to help you pinpoint the issues with your analytical algorithm. As it stands we see a bunch of red cells, but no obvious clues as to what exactly is causing its error rating.

For reference, an example of what I mean by compound GPS heat map:

It matters how the data were collected, i.e. its legal conditions.